2-INF-185 Integrácia dátových zdrojov 2018/19

Materiály · Úvod · Pravidlá · Kontakt
· Od 14.3. sa presúvame do učebne F2-T3, a.k.a. F2-128.
· Body z už opravených úloh ako aj komentáre k návrhu projektov nájdete na serveri v /grades/userid.txt
· Dátumy odovzdania projektov:
1. termín: streda 29.5. 22:00
2. termín: nedeľa 16.6. 22:00
Oba termíny sú riadne, prvý je určený pre študentov, ktorí chcú predmet ukončiť skôr. Pár dní po odvzdaní sa budú konať krátke osobné stretnutia s vyučujúcimi (diskusia k projektu a uzatváranie známky). Presné dni a časy dohodneme neskôr. Projekty odovzdajte podobne ako domáce úlohy do /submit/projekt


Integrácia dátových zdrojov 2017/18

From IDZ
Jump to: navigation, search

Website for 2017/18

Contents

Kontakt

Vyučujúci

Rozvrh

  • Štvrtok 14:50-17:10 F1-248

Úvod

Cieľová skupina

Tento predmet je určený pre študentov 2. ročníka bakalárskeho študijného programu Bioinformatika a pre študentov bakalárskeho a magisterského študijného programu Informatika, obzvlášť ak plánujú na magisterskom štúdiu absolvovať štátnicové zameranie Bioinformatika a strojové učenie. Radi privítame aj študentov iných zameraní a študijných programov, pokiaľ majú požadované (neformálne) prerekvizity.

Predpokladáme, že študenti na tomto predmete už vedia programovať v niektorom programovacom jazyku a neboja sa učiť podľa potreby nové jazyky. Takisto predpokladáme základnú znalosť práce v Linuxe vrátane spúšťania príkazov na príkazovom riadku (mali by ste poznať aspoň základné príkazy na prácu so súbormi a adresármi ako cd, mkdir, cp, mv, rm, chmod a pod.). Hoci väčšina technológií preberaných na tomto predmete sa dá použiť na spracovanie dát z mnohých oblastí, budeme ich často ilustrovať na príkladoch z oblasti bioinformatiky. Pokúsime sa vysvetliť potrebné pojmy, ale bolo by dobre, ak by ste sa orientovali v základných pojmoch molekulárnej biológie, ako sú DNA, RNA, proteín, gén, genóm, evolúcia, fylogenetický strom a pod. Študentom zamerania Bioinformatika a strojové učenie odporúčame absolvovať najskôr Metódy v bioinformatike, až potom tento predmet.

Ak sa chcete doučiť základy používania príkazového riadku, skúste napr. tento tutoriál: http://korflab.ucdavis.edu/bootcamp.html

Cieľ predmetu

Počas štúdia sa naučíte mnohé zaujímave algoritmy, modely a metódy, ktoré sa dajú použiť na spracovanie dát v bioinformatike alebo iných oblastiach. Ak však počas štúdia alebo aj neskôr v zamestnaní budete chcieť tieto metódy použiť na reálne dáta, zistíte, že väčšinou treba vynaložiť značné úsilie na samotné získanie dát, ich predspracovanie do vhodného tvaru, testovanie a porovnávanie rôznych metód alebo ich nastavení a získavanie finálnych výsledkov v tvare prehľadných tabuliek a grafov. Často je potrebné tieto činnosti veľakrát opakovať pre rôzne vstupy, rôzne nastavenia a podobne. Obzvlášť v bioinformatike je možné si nájsť zamestnanie, kde vašou hlavnou náplňou bude spracovanie dát s použitím už hotových nástrojov, prípadne doplnených menšími vlastnými programami. Na tomto predmete si ukážeme niektoré programovacie jazyky, postupy a technológie vhodné na tieto činnosti. Veľa z nich je použiteľných na dáta z rôznych oblastí, ale budeme sa venovať aj špecificky bioinformatickým nástrojom.

Základné princípy

Odporúčame nasledujúci článok s dobrými radami k výpočtovým experimentom:

Niektoré dôležité zásady:

  • Citát z článku Noble 2009: "Everything you do, you will probably have to do over again."
  • Dobre zdokumentujte všetky kroky experimentu (čo ste robili, prečo ste to robili, čo vám vyšlo)
    • Ani vy sami si o pár mesiacov tieto detaily nebudete pamätať
  • Snažte sa udržiavať logickú štruktúru adresárov a súborov
    • Ak však máte veľa experimentov, môže byť dostačujúce označiť ich dátumom, nevymýšľať stále nové dlhé mená
  • Snažte sa vyhýbať manuálnym úpravám medzivýsledkov, ktoré znemožňujú jednoduché zopakovanie experimentu
  • Snažte sa detegovať chyby v dátach
    • Skripty by mali skončiť s chybovou hláškou, keď niečo nejde ako by malo
    • V skriptoch čo najviac kontrolujte, že vstupné dáta zodpovedajú vašim predstavám (správny formát, rozumný rozsah hodnôt atď.)
    • Ak v skripte voláte iný program, kontrolujte jeho exit code
    • Tiež čo najčastejšie kontrolujte medzivýsledky výpočtu (ručným prezeraním, výpočtom rôznych štatistík a pod.), aby ste odhalili prípadné chyby v dátach alebo vo vašom kóde

Pravidlá

Známkovanie

  • Domáce úlohy: 55%
  • Návrh projektu: 5%
  • Projekt: 40%

Stupnica:

  • A: 90 a viac, B:80...89, C: 70...79, D: 60...69, E: 50...59, FX: menej ako 50%

Formát predmetu

  • Každý týždeň 3 vyučovacie hodiny, z toho cca prvá je prednáška a na ďalšie dve cvičenia. Na cvičeniach samostatne riešite príklady, ktoré doma dokončíte ako domácu úlohu.
  • Cez skúškové obdobie budete odovzdávať projekt. Po odovzdaní projektov sa bude konať ešte diskusia o projekte s vyučujúcimi, ktorá môže ovplyvniť vaše body z projektu.
  • Budete mať konto na Linuxovom serveri určenom pre tento predmet. Toto konto používajte len na účely tohto predmetu a snažte sa server príliš svojou aktivitou nepreťažiť, aby slúžil všetkým študentom. Akékoľvek pokusy úmyselne narušiť chod servera budú považované za vážne porušenie pravidiel predmetu.

Domáce úlohy

  • Termín DÚ týkajúcej sa aktuálnej prednášky je vždy do 9:00 v deň nasledujúcej prednášky (t.j. väčšinou o necelý týždeň od zadania).
  • Domácu úlohu odporúčame začať robiť na cvičení, kde vám môžeme prípadne poradiť. Ak máte otázky neskôr, pýtajte sa vyučujúcich emailom.
  • Domácu úlohu môžete robiť na ľubovoľnom počítači, pokiaľ možno pod Linuxom. Odovzdaný kód alebo príkazy by však mali byť spustiteľné na serveri pre tento predmet, nepoužívajte teda špeciálny softvér alebo nastavenia vášho počítača.
  • Domáca úloha sa odovzdáva nakopírovaním požadovaných súborov do požadovaného adresára na serveri. Konkrétne požiadavky budú spresnené v zadaní.
  • Ak sú mená súborov špecifikované v zadaní, dodržujte ich. Ak ich vymýšľate sami, nazvite ich rozumne. V prípade potreby si spravte aj podadresáre, napr. na jednotlivé príklady.
  • Dbajte na prehľadnosť odovzdaného zdrojového kódu (odsadzovanie, rozumné názvy premenných, podľa potreby komentáre)

Protokoly

  • Väčšinou bude požadovanou súčasťou úlohy textový dokument nazvaný protokol.
  • Protokol môže byť vo formáte .txt alebo .pdf a jeho meno má byť protocol.pdf alebo protocol.txt (nakopírujte ho do odovzdaného adresára)
  • Protokol môže byť po slovensky alebo po anglicky.
  • V prípade použitia txt formátu a diakritiky ju kódujte v UTF8, ale pre jednoduchosť môžete protokoly písať aj bez diakritiky. Ak je protocol v pdf formáte, mali by sa v ňom dať selektovať texty.
  • Vo väčšine úloh dostanete kostru protokolu, dodržujte ju.

Hlavička protokolu, vyhodnotenie

  • Na vrchu protokolu uveďte meno, číslo domácej úluhy a vaše vyhodnotenie toho, ako sa vám úlohu podarilo vyriešiť. Vyhodnotenie je prehľadný zoznam všetkých príkladov zo zadania, ktoré ste aspoň začali riešiť a kódov označujúcich ich stupeň dokončenia:
    • kód HOTOVO uveďte, ak si myslíte, že tento príklad máte úplne a správne vyriešený
    • kód ČASŤ uveďte, ak ste nevyriešili príklad celý a do poznámky za kód stručne uveďte, čo máte hotové a čo nie, prípadne ktorými časťami si nie ste istí.
    • kód MOŽNO uveďte, ak príklad máte celý, ale nie ste si istí, či správne. Opäť v poznámke uveďte, čím si nie ste istí.
    • kód NIČ uveďte, ak ste príklad ani nezačali riešiť
  • Vaše vyhodnotenie je pre nás pomôckou pri bodovaní. Príklady označené HOTOVO budeme kontrolovať námatkovo, k príkladom označeným MOŽNO sa vám pokúsime dať nejakú spätnú väzbu, takisto aj k príkladom označeným ČASŤ, kde v poznámke vyjadríte, že ste mali nejaké problémy.
  • Pri vyhodnotení sa pokúste čo najlepšie posúdiť správnosť vašich riešení, pričom kvalita vášho seba-hodnotenia môže vplývať na celkový počet bodov.

Obsah protokolu

  • Ak nie je v zadaní určené inak, protokol by mal obsahovať nasledovné údaje:
    • Zoznam odovzdaných súborov: o každom súbore uveďte jeho význam a či ste ho vyrobili ručne, získali z externých zdrojov alebo vypočítali nejakým programom. Ak máte väčšie množstvo súborov so systematickým pomenovaním, stačí vysvetliť schému názvov všeobecne. Súbory, ktorých mená sú špecifikované v zadaní, nemusíte v zozname uvádzať.
    • Postupnosť všetkých spustených príkazov, prípadne iných krokov, ktorými ste dospeli k získaným výsledkom. Tu uvádzajte príkazy na spracovanie dát a spúšťanie vašich či iných programov. Netreba uvádzať príkazy súvisiace so samotným programovaním (spúšťanie editora, nastavenie práv na spustenie a pod.), s kopírovaním úlohy na server a pod. Uveďte aj stručné komentáre, čo bolo účelom určitého príkazu alebo skupiny príkazov.
    • Zoznam zdrojov: webstránky a pod., ktoré ste pri riešení úlohy použili. Nemusíte uvádzať webstránku predmetu a zdroje odporučené priamo v zadaní.

Celkovo by protokol mal umožniť čitateľovi zorientovať sa vo vašich súboroch a tiež v prípade záujmu vykonať rovnaké výpočty, akými ste dospeli vy k výsledku. Nemusíte písať slohy, stačia zrozumiteľné a prehľadné heslovité poznámky.

Projekty

Cieľom projektu je vyskúšať si naučené zručnosti na konkrétnom projekte spracovania dát. Vašou úlohou je zohnať si dáta, tieto dáta analyzovať niektorými technikami z prednášok, prípadne aj inými technológiami a získané výsledky zobraziť v prehľadných grafoch a tabuľkách. Ideálne je, ak sa vám podarí prísť k zaujímavým alebo užitočným záverom, ale hodnotiť budeme hlavne voľbu vhodného postupu a jeho technickú náročnosť. Rozsah samotného programovania alebo analýzy dát by mal zodpovedať zhruba dvom domácim úlohám, ale celkovo bude projekt náročnejší, lebo na rozdiel od úloh nemáte postup a dáta vopred určené, ale musíte si ich vymyslieť sami a nie vždy sa prvý nápad ukáže ako správny. V projekte môžete využiť aj existujúce nástroje a knižnice, ale pokiaľ možno používajte nástroje spúšťané na príkazovom riadku.

Zhruba v dvoch tretinách semestra budete odovzdávať návrh projektu (formát txt alebo pdf, rozsah 0.5-1 strana). V tomto návrhu uveďte, aké dáta budete spracovávať, ako ich zoženiete, čo je cieľom analýzy a aké technológie plánujete použiť. Ciele a technológie môžete počas práce na projekte mierne pozmeniť podľa okolností, mali by ste však mať počiatočnú predstavu. K návrhu vám dáme spätnú väzbu, pričom v niektorých prípadoch môže byť potrebné tému mierne alebo úplne zmeniť. Za načas odovzdaný vhodný návrh projektu získate 5% z celkovej známky. Návrh odporúčame pred odovzdaním konzultovať s vyučujúcimi.

Cez skúškové obdobie bude určený termín odovzdania projektu. Podobne ako pri domácich úlohách odovzdávajte adresár s požadovanými súbormi:

  • Vaše programy a súbory s dátami (veľmi veľké dátové súbory vynechajte)
  • Protokol podobne ako pri domácich úlohách
    • formát txt alebo pdf, stručné heslovité poznámky
    • obsahuje zoznam súborov, podrobný postup pri analýze dát (spustené príkazy), ako aj použité zdroje (dáta, programy, dokumentácia a iná literatúra atď)
  • Správu k projektu vo formáte pdf. Na rozdiel od menej formálneho protokolu by správu mal tvoriť súvislý text v odbornom štýle, podobne ako napr. záverečné práce. Môžete písať po slovensky alebo po anglicky, ale pokiaľ možno gramaticky správne. Správa by mala mať tieto časti:
    • úvod, v ktorom vysvetlíte ciele projektu, prípadne potrebné poznatky zo skúmanej oblasti a aké dáta ste mali k dispozícii
    • stručný popis metód, v ktorom neuvádzajte detailne jednotlivé kroky, skôr prehľad použitého prístupu a jeho zdôvodnenie
    • výsledky analýzy (tabuľky, grafy a pod.) a popis týchto výsledkov, prípadne aké závery sa z nich dajú spraviť (nezabudnite vysvetliť, čo znamenajú údaje v tabuľkách, osi grafov a pod.). Okrem finálnych výsledkov analýzy uveďte aj čiastkové výsledky, ktorými ste sa snažili overovať, že pôvodné dáta a jednotlivé časti vášho postupu sa správajú rozumne.
    • diskusiu, v ktorej uvediete, ktoré časti projektu boli náročné a na aké problémy ste narazili, kde sa vám naopak podarilo nájsť spôsob, ako problém vyriešiť jednoducho, ktoré časti projektu by ste spätne odporúčali robiť iným než vašim postupom, čo ste sa na projekte naučili a podobne

Projekty môžete robiť aj vo dvojici, vtedy však vyžadujeme rozsiahlejší projekt a každý člen by mal byť primárne zodpovedný za určitú časť projektu, čo uveďte aj v správe. Dvojice odovzdávajú jednu správu, ale po odovzdaní projektu majú stretnutie s vyučujúcimi individuálne.

Ako nájsť tému projektu:

  • Môžete spracovať nejaké dáta, ktoré potrebujete do bakalárskej alebo diplomovej práce, prípadne aj dáta, ktoré potrebujte na iný predmet (v tom prípade uveďte v správe, o aký predmet ide a takisto upovedomte aj druhého vyučujúceho, že ste použili spracovanie dát ako projekt pre tento predmet). Obzvlášť pre BIN študentov môže byť tento predmet vhodnou príležitosťou nájsť si tému bakalárskej práce a začať na nej pracovať.
  • Môžete skúsiť zopakovať analýzu spravenú v nejakom vedeckom článku a overiť, že dostanete tie isté výsledky. Vhodné je tiež skúsiť analýzu aj mierne obmeniť (spustiť na iné dáta, zmeniť nejaké nastavenia, zostaviť aj iný typ grafu a pod.)
  • Môžete skúsiť nájsť niekoho, kto má dáta, ktoré by potreboval spracovať, ale nevie ako na to (môže ísť o biológov, vedcov z iných oblastí, ale aj neziskové organizácie a pod.) V prípade, že takýmto spôsobom kontaktujete tretie osoby, bolo by vhodné pracovať na projekte obzvlášť zodpovedne, aby ste nerobili zlé meno našej fakulte.
  • V projekte môžete porovnávať niekoľko programov na tú istú úlohu z hľadiska ich rýchlosti či presnosti výsledkov. Obsahom projektu bude príprava dát, na ktorých budete programy bežať, samotné spúšťanie (vhodne zoskriptované) ako aj vyhodnotenie výsledkov.
  • A samozrejme môžete niekde na internete vyhrabať zaujímavé dáta a snažiť sa z nich niečo vydolovať.

Opisovanie

  • Máte povolené sa so spolužiakmi a ďalšími osobami rozprávať o domácich úlohách resp. projektoch a stratégiách na ich riešenie. Kód, získané výsledky aj text, ktorý odovzdáte, musí však byť vaša samostatná práca. Je zakázané ukazovať svoj kód alebo texty spolužiakom.
  • Pri riešení domácej úlohy a projektu očakávame, že budete využívať internetové zdroje, najmä rôzne manuály a diskusné fóra k preberaným technológiám. Nesnažte sa však nájsť hotové riešenia zadaných úloh. Všetky použité zdroje uveďte v domácich úlohách a projektoch.
  • Ak nájdeme prípady opisovania alebo nepovolených pomôcok, všetci zúčastnení študenti získajú za príslušnú domácu úlohu, projekt a pod. nula bodov (t.j. aj tí, ktorí dali spolužiakom odpísať) a prípad ďalej podstúpime na riešenie disciplinárnej komisii fakulty.

Zverejňovanie

Zadania a materiály k predmetu sú voľne prístupné na tejto stránke. Prosím vás ale, aby ste nezverejňovali ani inak nešírili vaše riešenia domácich úloh, ak nie je v zadaní povedané inak. Vaše projekty môžete zverejniť, pokiaľ to nie je v rozpore s vašou dohodou so zadávateľom projektu a poskytovateľom dát.

L01

Lecture 1: Perl, part 1

Why Perl

  • From Wikipedia: It has been nicknamed "the Swiss Army chainsaw of scripting languages" because of its flexibility and power, and possibly also because of its "ugliness".

Oficial slogans:

  • There's more than one way to do it
  • Easy things should be easy and hard things should be possible

Advantages

  • Good capabilities for processing text files, regular expressions, running external programs etc.
  • Closer to common programming language than shell scripts
  • Perl one-liners on the command line can replace many other tools such as sed and awk
  • Many existing libraries

Disadvantages

  • Quirky syntax
  • It is easy to write very unreadable programs (sometimes joking called write-only language)
  • Quite slow and uses a lot of memory. If possible do no read entire input to memory, process line by line

Warning: we will use Perl 5, Perl 6 is quite a different language

Sources of Perl-related information

  • In package perl-doc man pages:
    • man perlintro introduction to Perl
    • man perlfunc list of standard functions in Perl
    • perldoc -f split describes function split, similarly other functions
    • perldoc -q sort shows answers to commonly asked questions (FAQ)
    • man perlretut and man perlre regular expressions
    • man perl list of other manual pages about Perl
  • The same content on the web http://perldoc.perl.org/
  • Various web tutorials e.g. this one
  • Books
    • Simon Cozens: Beginning Perl [1] freely downloadable
    • Larry Wall et al: Programming Perl [2] classics, Camel book
  • Bioperl [3] big library for bioinformatics
  • Perl for Windows: http://strawberryperl.com/

Hello world

It is possible to run the code directly from a command line (more later):

perl -e'print "Hello world\n"'

This is equivalent to the following code stored in a file:

#! /usr/bin/perl -w
use strict;
print "Hello world!\n";
  • First line is a path to the interpreter
  • Swith -w switches warnings on, e.g. if we manipulate with an undefined value (equivalen to "use warnings;")
  • Second line use strict will switch on a more strict syntax checks, e.g. all variables must be defined
  • Use of -w and use strict is strongly recommended
  • Store the program in a file, e.g. hello.pl
  • Make it executable (chmod a+x hello.pl)
  • Run it with command ./hello.pl
  • Also possible to run as perl hello.pl (e.g. if we don't have the path to the interpreter in the file or the executable bit set)

The first input file for today: sequence repeats

  • In genomes some sequences occur in many copies (often not exactly equal, only similar)
  • We have downloaded a table containing such sequence repeats on chromosome 2L of the fruitfly Drosophila melanogaster
  • It was done as follows: on webpage http://genome.ucsc.edu/ we select drosophila genome, then in main menu select Tools, Table browser, select group: variation and repeats, track: ReapatMasker, region: position chr2L, output format: all fields from the selected table a output file: repeats.txt
  • Each line of the file contains data about one repeat in the selected chromosome. The first line contains column names. Columns are tab-separated. Here are the first two lines:
#bin    swScore milliDiv        milliDel        milliIns        genoName        genoStart       genoEnd genoLeft        strand  repName repClass        repFamily       repStart        repEnd  repLeft id
585     778     167     7       20      chr2L   1       154     -23513558       +       HETRP_DM        Satellite       Satellite       1519    1669    -203    1
  • The file can be found at our server under filename /tasks/hw01/repeats.txt (17185 lines)
  • A small randomly selected subset of the table rows is in file /tasks/hw01/repeats-small.txt (159 lines)

A sample Perl program

For each type of repeat (column 11 of the file when counting from 0) we want to compute the number of repeats of this type

#!/usr/bin/perl -w
use strict;

#associative array (hash), with repeat type as key
my %count;  

while(my $line = <STDIN>) {  # read every line on input
    chomp $line;    # delete end of line, if any

    if($line =~ /^#/) {  # skip commented lines
       next;       # similar to "continue" in C, move to next iteration
    }

    # split the input line to columns on every tab, store them in an array
    my @columns = split "\t", $line;  

    # check input - should have at least 17 columns
    die "Bad input '$line'" unless @columns >= 17;

    my $type = $columns[11];

    # increase counter for this type
    $count{$type}++;
}

# write out results, types sorted alphabetically
foreach my $type (sort keys %count) {
    print $type, " ", $count{$type}, "\n";
}

This program does the same thing as the following one-liner (more on one-liners in two weeks)

perl -F'"\t"' -lane 'next if /^#/; die unless @F>=17; $count{$F[11]}++; END { foreach (sort keys %count) { print "$_ $count{$_}" }}' filename


The second input file for today: DNA sequencing reads (fastq)

  • DNA sequencing machines can read only short pieces of DNA called reads
  • Reads are usually stored in fastq format
  • Files can be very large (gigabytes or more), but we will use only a small sample from bacteria Staphylococcus aureus, source [4]
  • Each read is on 4 lines:
    • line 1: ID of the read and other description, line starts with @
    • line 2: DNA sequence, A,C,G,T are bases (nucleotides) of DNA, N means unknown base
    • line 3: +
    • line 4: quality string, which is the string of the same length as DNA in line 2. Each character represents quality of one base in DNA. If p is the probability that this base is wrong, the quality string will contain character with ASCII value 33+(-10 log p), where log is decimal logarithm. This means that higher ASCII means base of higher quality. Character ! (ASCII 33) means probability 1 of error, character $ (ASCII 36) means 50% error, character + (ASCII 43) is 10% error, character 5 (ASCII 53) is 1% error.
    • Note that some sequencing platforms represent qualities differently (see article linked above)
  • Our file has all reads of equal length (this is not always the case)
  • Technically, a single read and its quality can be split into multiple lines, but this is rarely done and we will assume that each read takes 4 lines as described above

The first 4 reads from file /tasks/hw01/reads-small.fastq

@SRR022868.1845/1
AAATTTAGGAAAAGATGATTTAGCAACATTTAGCCTTAATGAAAGACCAGATTCTGTTGCCATGTTTGAATGCCTTAAACCAGTAGCAGAATCAGTATAAA
+
IICIIIIIIIIIID%IIII8>I8III1II,II)I+III*II<II,E;-HI>+I0IB99I%%2GI*=?5*&1>'$0;%'+%%+;#'$&'%%$-+*$--*+(%
@SRR022868.1846/1
TAGCGTTGTAAAATAAATTTCTAGAATGGAAGTGATGATATTGAAATACACTCAGATCCTGAATGAAAGATTTATTAAAGTTAAGACGAGAGTCTCATTAT
+
4CIIIIIIII52I)IIIII0I16IIIII2IIII;IIAII&I6AI+*+&G5&G.@8/6&%&,03:*.$479.91(9--$,*&/3"$#&*'+#&##&$(&+&+

Read the rest of the lecture on your own as you need for #HW01


Variables, types

Scalar variables

  • Scalar variables start with $, they can hold undefined value (undef), string, number, reference etc.
  • Perl converts automatically between strings and numbers
perl -e'print((1 . "2")+1, "\n")'
13
perl -e'print(("a" . "2")+1, "\n")'
1
perl -we'print(("a" . "2")+1, "\n")'
Argument "a2" isn't numeric in addition (+) at -e line 1.
1
  • If we switch on strict parsing, each variable needs to be defined by my, several variables created and initialized as follows: my ($a,$b) = (0,1);
  • Usual set of C-style operators, power is **, string concatenation .
  • Numbers compared by <, <=, ==, != etc., strings by lt, le, eq, ne, gt, ge,
  • Comparison operator $a cmp $b for strings, $a <=> $b for numbers: returns -1 if $a<$b, 0 if they are equal, +1 if $a>$b

Arrays

  • Names start with @, e.g. @a
  • Access to element 0 in array: $a[0]
    • Starts with $, because the expression as a whole is a scalar value
  • Length of array scalar(@a). In scalar context, @a is the same thing.
    • e.g. for(my $i=0; $i<@a; $i++) { ... }
  • If using non-existent indexes, they will be created, initialized to undef (++, += treat undef as 0)
  • Stack/vector using functions push and pop: push @a, (1,2,3); $x = pop @a;
  • Analogicaly shift and unshift on the left end of the array (slower)
  • Sorting
    • @a = sort @a; (sorts alphabetically)
    • @a = sort {$a <=> $b} @a; (sort numerically)
    • { } can contain arbitrary comparison function, $a and $b are the two compared elements
  • Array concatenation @c = (@a,@b);
  • Swap values of two variables: ($x,$y) = ($y,$x);
  • Iterate through values of an array (values can be changed):
perl -e'my @a = (1,2,3); foreach my $val (@a) { $val++; } print join(" ", @a), "\n";'
2 3 4

Associative array (hashes)

  • Names start with %, e.g. %b
  • Access element with name "X": $b{"X"}
  • Write out all elements of associative array %b
foreach my $key (keys %b) {
  print $key, " ", $b{$key}, "\n";
}
  • Initialization with constant: %b = ("key1"=>"value1","key2"=>"value2")
    • instead of => you can also use ,
  • test for existence of a key: if(exists $a{"x"}) {...}
  • (other methods will create the queried key with undef value)

Multidimensional arrays, fun with pointers

  • Pointer to a variable: \$a, \@a, \%a
  • Pointer to an anonymous array: [1,2,3], pointer to an anonymous hash: {"kluc1"=>"hodnota1"}
  • Hash of lists:
my %a = ("fruits"=>["apple","banana","orange"], "vegetables"=>["celery","carrot"]}
$x = $a{"fruits"}[1];
push @{$a{"fruits"}}, "kiwi";
my $aref = \%a;
$x = $aref->{"fruits"}[1];
  • Module Data::Dumper has function Dumper, which will recursively print complex data structures

Strings, regular expressions

Strings

  • Substring: substr($string, $start, $length)
    • used also to access individual charaters (use length 1)
    • If we omit $length, considers until the end of the string, negative start counted from the end of the stringzaciatok rata od konca,...
    • We can also used to replace a substring by something else: substr($str, 0, 1) = "aaa" (replaces the first character by "aaa")
  • Length of a string: length($str)
  • Splitting a string to parts: split reg_expression, $string, $max_number_of_parts
    • if " " instead of regular expression, splits at whitespace
  • Connecting parts join($separator, @strings)
  • Other useful functions: chomp (removes end of line), index (finds a substring), lc, uc (conversion to lowercase/uppercase), reverse (mirror image), sprintf (C-style formatting)

Regular expressions

$line =~ s/\s+$//;   # remove whitespace at the end of the line
$line =~ s/[0-9]+/X/g;  # replace each sequence of numbers with character X

#from the name of the fasta sequence (starting with >) create a string until the first space
#(\S means non-whitespace), the result is stored in $1, as specified by ()
if($line =~ /^\>(\S+)/) { $name = $1; }

perl -le'$X="123 4 567"; $X=~s/[0-9]+/X/g; print $X'
X X X

Conditionals, loops

if(expression) {  # [] and () cannot be omitted
   commands
} elsif(expression) {
   commands
} else {
   commands
}

command if expression;   # here () not necessary
command unless expression;
die "negative value of x: $x" unless $x>=0;

for(my $i=0; $i<100; $i++) {
   print $i, "\n";
}

foreach my $i (0..99) {
   print $i, "\n";
}

$x=1;
while(1) {
   $x *= 2;
   last if $x>=100;
}
  • Undefined value, number 0 and strings "" and "0" evaluate as false, but I would recommmend always explicitly using logical values in conditional expressions, e.g. if(defined $x), if($x eq ""), if($x==0) etc.

Input, output

  • Reading one line from standard input: $line = <STDIN>
  • If no more input data available, returns undef
  • See also [6]
  • Special idiom while(my $line = <STDIN>) equivalent to while (defined(my $line = <STDIN>))
    • iterates through all lines of input
  • chomp $line removes "\n", if any from the end of the string
  • output to stdout through print or printf

HW01

See Lecture 1

Files

We have 4 input files for this homework. We recommend creating soft links to your working directory as follows:

ln -s /tasks/hw01/repeats-small.txt .  # small version of the repeat file
ln -s /tasks/hw01/repeats.txt .        # full version of the repeat file
ln -s /tasks/hw01/reads-small.fastq .  # smaller version of the read file
ln -s /tasks/hw01/reads.fastq .        # bigger version of the read file

We recommend writing your protocol starting from an outline provided in /tasks/hw01/protocol.txt

Submitting

  • Directory /submit/hw01/your_username will be created for you
  • Copy required files to this directory, including the protocol named protocol.txt or protocol.pdf
  • You can modify these files freely until deadline, but after the deadline of the homework, you will lose access rights to this directory

Task A

  • Consider the program for counting repeat types in the lecture 1, save it to file repeat-stat.pl
  • Extend it to compute the average length of each type of repeat
    • Each row of the input table contains the start and end coordinates of the repeat in columns 7 and 6. The length is simply the difference of these two values.
  • Output a table with three columns: type of repeat, the number of occurrences, the average length of the repeat.
    • Use printf to print these three items right-justified in columns of sufficient width, print the average length to 1 decimal place.
  • If you run your script on the small file, the output should look something like this (exact column widths may differ):
./repeat-stat.pl < repeats-small.txt
                 DNA         5     377.4
                LINE         4     410.2
                 LTR        13     355.4
      Low_complexity        22      47.2
                  RC         8     236.2
       Simple_repeat       106      39.0
  • Include in your protocol the output when you run your script on the large file: ./repeat-stat.pl < repeats.txt
  • Find out on Wikipedia, what acronyms LINE and LTR stand for. Do their names correspond to their lengths? (Write a short answer in the protocol.)
  • Submit only your script, repeat-stat.pl

Task B

  • Write a script which reformats FASTQ file to FASTA format, call it fastq2fasta.pl
    • fastq file should be on standard input, fasta file written to standard output
  • FASTA format is a typical format for storing DNA and protein sequences.
    • Each sequence consists of several lines of the file. The first line starts with ">" followed by identifier of the sequence and optionally some further description separated by whitespace
    • The sequence itself is on the second line, long sequences are split into multiple lines
  • In our case, the name of the sequence will be the ID of the read with @ replaced by > and / replaced by _
  • For example, the first two reads of reads.fastq are:
@SRR022868.1845/1
AAATTTAGGAAAAGATGATTTAGCAACATTTAGCCTTAATGAAAGACCAGATTCTGTTGCCATGTTTGAATGCCTTAAACCAGTAGCAGAATCAGTATAAA
+
IICIIIIIIIIIID%IIII8>I8III1II,II)I+III*II<II,E;-HI>+I0IB99I%%2GI*=?5*&1>'$0;%'+%%+;#'$&'%%$-+*$--*+(%
@SRR022868.1846/1
TAGCGTTGTAAAATAAATTTCTAGAATGGAAGTGATGATATTGAAATACACTCAGATCCTGAATGAAAGATTTATTAAAGTTAAGACGAGAGTCTCATTAT
+
4CIIIIIIII52I)IIIII0I16IIIII2IIII;IIAII&I6AI+*+&G5&G.@8/6&%&,03:*.$479.91(9--$,*&/3"$#&*'+#&##&$(&+&+
  • These should be reformatted as follows:
>SRR022868.1845_1
AAATTTAGGAAAAGATGATTTAGCAACATTTAGCCTTAATGAAAGACCAGATTCTGTTGCCATGTTTGAATGCCTTAAACCAGTAGCAGAATCAGTATAAA
>SRR022868.1846_1
TAGCGTTGTAAAATAAATTTCTAGAATGGAAGTGATGATATTGAAATACACTCAGATCCTGAATGAAAGATTTATTAAAGTTAAGACGAGAGTCTCATTAT
  • Submit files fastq2fasta.pl and reads-small.fasta
    • the latter file is created by running ./fastq2fasta.pl < reads-small.fastq > reads-small.fasta

Task C

  • Write a script fastq-quality.pl which for each position in a read computes the average quality
  • Standard input has fastq file with multiple reads, possibly of different lengths
  • As quality we will use ASCII values of characters in the quality string with value 33 subtracted, so the quality is -10 log p
    • ASCII value can be computed by function ord
  • Positions in reads will be numbered from 0
  • Since reads can differ in length, some positions are used in more reads, some in fewer
  • For each position from 0 up to the highest position used in some read, print three numbers separated by tabs "\t": the position index, the number of times this position was used in reads, average quality at that position with 1 decimal place (you can again use printf)
  • The last two lines when you run ./fastq-quality.pl < reads-small.fastq should be
99      86      5.5
100     86      8.6
  • Run the following command, which runs your script on the larger file and selects every 10th position. Include the output in your protocol. Do you see any trend in quality values with increasing position? (Include a short comment in protocol.)
./fastq-quality.pl < reads.fastq | perl -lane 'print if $F[0]%10==0'
  • Submit only fastq-quality.pl

Task D

  • Write script fastq-trim.pl that trims low quality bases from the end of each read and filters out short reads
  • This script should read a fastq file from standard input and write trimmed fastq file to standard output
  • It should also accept two command-line arguments: character Q and integer L
    • We have not covered processing command line arguments, but you can use the code snippet below
  • Q is the minimum acceptable quality (characters from quality string with ASCII value >= ASCII value of Q are ok)
  • L is the minimum acceptable length of a read
  • First find the last base in a read which has quality at least Q (if any). All bases after this base will be removed from both the sequence and quality string
  • If the resulting read has fewer than L bases, it is omitted from the output

You can check your program by the following tests:

  • If you run the following two commands, you should get tmp identical with input and thus output of diff should be empty
./fastq-trim.pl '!' 101 < reads-small.fastq > tmp  # trim at quality ASCII >=33 and length >=101
diff reads-small.fastq tmp                         # output should be empty (no differences)
  • If you run the following two commands, you should see differences in 4 reads, 2 bases trimmed from each
./fastq-trim.pl '"' 1 < reads-small.fastq > tmp   # trim at quality ASCII >=34 and length >=1
diff reads-small.fastq tmp                        # output should be differences in 4 reads
  • If you run the following commands, you should get empty output (no reads meet the criteria):
./fastq-trim.pl d 1 < reads-small.fastq           # quality ASCII >=100, length >= 1
./fastq-trim.pl '!' 102 < reads-small.fastq       # quality ASCII >=33 and length >=102

Further runs and submitting

  • Run ./fastq-trim.pl '(' 95 < reads-small.fastq > reads-small-filtered.fastq # quality ASCII >= 40
  • Submit files fastq-trim.pl and reads-small-filtered.fastq
  • If you have done task C, run quality statistics on the trimmed version of the bigger file using command below and include the result in the protocol. Comment in the protocol on differences between statistics on the whole file in part C and D. Are they as you expected?
./fastq-trim.pl 2 50 < reads.fastq | ./fastq-quality.pl | perl -lane 'print if $F[0]%10==0'  # quality ASCII >= 50
  • Note: you have created tools which can be combined, e.g. you can first trim fastq and then convert it to fasta (no need to submit these files)

Parsing command-line arguments in this task (they will be stored in variables $Q and $L):

#!/usr/bin/perl -w
use strict;

my $USAGE = "
Usage:
$0 Q L < input.fastq > output.fastq

Trim from the end of each read bases with ASCII quality value less
than the given threshold Q. If the length of the read after trimming
is less than L, the read will be omitted from output.

L is a non-negative integer, Q is a character
";

# check that we have exactly 2 command-line arguments
die $USAGE unless @ARGV==2;
# copy command-line arguments to variables Q and L
my ($Q, $L) = @ARGV;
# check that $Q is one character and $L looks like a non-negative integer
die $USAGE unless length($Q)==1 && $L=~/^[0-9]+$/;

L02

Motivation: Building Phylogenetic Trees

The task for today will be to build a phylogenetic tree of several species using sequences of several genes.

  • A phylogenetic tree is a tree showing evolutionary history of these species. Leaves are target present-day species, internal nodes are their common ancestors.
  • Input contains sequences of genes from each species.
  • Step 1: Identify ortholog groups. Orthologs are genes from different species that "correspond" to each other. This is done based on sequence similarity and we can use a tool called blast to identify sequence similarities between individual genes. The result of ortholog group identification will be a set of genes, each gene having one sequence from each of the 6 species
chimp_94013 dog_84719 human_15749 macaque_34640 mouse_17461 rat_09232
  • Step 2: For each ortholog group, we need to align genes and build a phylogenetic tree for this gene using existing methods. We can do this using tools muscle (for alignment) and phyml (for phylogenetic tree inference).

Unaligned sequences:

>mouse
ATGCAGTTCCCGCACCCGGGGCCCGCGGCTGCGCCCGCCGTGGGAGTCCCGCTGTATGCG
>rat
ATGCAGTTCCCGCACCCGGGGCCCGCGGCTGCGCCCGCCGTCGGAGTCCCGCTGTACGCG
>dog
ATGCAGTACCACCCCGGGCCGGCGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>human
ATGCAGTACCCGCACCCCGGGCCGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>chimp
ATGCAGTACCCGCACCCCGGGCCGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>macaque
ATGCAGTACCCGCACCCCGGGCGGCGGCCGTGGGGGTGGC

Aligned sequences:

>mouse
ATGCAGTTCCCGCACCCGGGGCCCGCGGCTGCGCCCGCCGTGGGAGTCCCGCTGTATGCG
>rat
ATGCAGTTCCCGCACCCGGGGCCCGCGGCTGCGCCCGCCGTCGGAGTCCCGCTGTACGCG
>dog
ATGCAGTAC---CACCCCGGGCCGGCGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>human
ATGCAGTACCCGCACCCCGGGC---CGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>chimp
ATGCAGTACCCGCACCCCGGGC---CGGCGGCGGGCGCCGTGGGGGTGCCGCTGTACGCG
>macaque
ATGCAGTACCCGCACCCCGGGC----------GGCGGCCGTGGGGGTGGC----------

Phylogenetic tree:

(mouse:0.03240286,rat:0.01544553,(dog:0.03632419,(macaque:0.01505050,(human:0.00000001,chimp:0.00000001):0.00627957):0.01396920):0.10645019);
Tree for gene human_15749 (branch lengths ignored)


  • Step 3: The result of the previous step will be several trees, one for every gene. Ideally, all trees would be identical, showing the real evolutionary history of the six species. But it is not easy to infer the real tree from sequence data, so trees from different genes might differ. Therefore, in the last step, we will build a consensus tree. This can be done by usina interactive tool called phylip.
  • Output is a single consensus tree.
Tree for gene human_15749
Tree for gene human_13531
Tree for gene human_31770
Strict consensus for the three gene trees


Our goal for today is to build a pipeline that automates the whole task.

Opening files

my $in;
open $in, "<", "path/file.txt" or die;  # open file for reading
while(my $line = <$in>) {
  # process line
}
close $in;

my $out;
open $out, ">", "path/file2.txt" or die; # open file for writing
print $out "Hello world\n";
close $out;
# if we want to append to a file use the following instead:
# open $out, ">>", "cesta/subor2.txt" or die;

# standard files
print STDERR "Hello world\n";
my $line = <STDIN>;
# files as arguments of a function
citaj_subor($in);
citaj_subor(\*STDIN);

Working with files and directories

Working directories or files with automatically generated names are automagically deleted after the program finishes.

use File::Temp qw/tempdir/;
my $dir = tempdir("atoms_XXXXXXX", TMPDIR => 1, CLEANUP => 1 ); 
print STDERR "Creating temporary directory $dir\n";
open $out,">$dir/myfile.txt" or die;

Copying files

use File::Copy;
copy("file1","file2") or die "Copy failed: $!";
copy("Copy.pm",\*STDOUT);
move("/dev1/fileA","/dev2/fileB");

Other functions for working with file system, e.g. chdir, mkdir, unlink, chmod, ...

Function glob finds files with wildcard characters similarly as on command line (see also opendir, readdir, and File::Find module)

ls *.pl
perl -le'foreach my $f (glob("*.pl")) { print $f; }'

Additional functions for working with file names, paths, etc. in modules File::Spec and File::Basename.

Testing for an existence of a file (more in perldoc -f -X)

if(-r "file.txt") { ... }  # is file.txt readable?
if(-d "dir") {.... }       # is dir a directory?

Running external programs

my $ret = system("command arguments");
# returns -1 if cannot run command, otherwise pass the return code
my $allfiles = `ls`;
# returns the result of a command as a text
# cannot test return code

Using pipes

open $in, "ls |";
while(my $line = <$in>) { ... }
open $out, "| wc"; 
print $out "1234\n"; 
close $out;'

      1       1       5

Command-line arguments

# module for processing options in a standardized way
use Getopt::Std;
# string with usage manual
my $USAGE = "$0 [options] length filename

Options:
-l           switch on lucky mode
-o filename  write output to filename
";

# all arguments to the command are stored in @ARGV array
# parse options and remove them from @ARGV
my %options;
getopts("lo:", \%options);
# now there should be exactly two arguments in @ARGV
die $USAGE unless @ARGV==2;
# process options
my ($length, $filenamefile) = @ARGV;
# values of options are in the %options array
if(exists $options{'l'}) { print "Lucky mode\n"; }

For long option names, see module Getopt::Long

Defining functions

Defining new functions

sub function_name {
  # arguments are stored in @_ array
  my ($firstarg, $secondarg) = @_;
  # do something
  return ($result, $second_result);
}
  • Arrays and hashes are usually passed as references: function_name(\@array, \%hash);
  • It is advantageous to pass long string as references as well to prevent needless copying: function_name(\$sequence);
  • References need to be dereferenced, e.g. substr($$sequence) or $array->[0]

Bioperl

use Bio::Tools::CodonTable;
sub translate
{
    my ($seq, $code) = @_;
    my $CodonTable = Bio::Tools::CodonTable->new( -id => $code);
    my $result = $CodonTable->translate($seq);

    return $result;
}


Defining modules

Module with name XXX should be in file XXX.pm.

package shared;

BEGIN {
    use Exporter   ();
    our (@ISA, @EXPORT, @EXPORT_OK);
    @ISA = qw(Exporter);
    # symbols to export by default
    @EXPORT = qw(funkcia1, funkcia2);
}

sub funkcia1 {
...
}

sub funkcia2 {
...
}

#module must return true
1;

Using the module located in the same directory as .pl file:

use FindBin qw($Bin);  # $Bin is the directory with the script
use lib "$Bin";        # add bin to the library path
use shared;

HW02

Biological background and overall approach

The task for today will be to build a phylogenetic tree of several species using sequences of several genes.

  • We will use 6 mammals: human, chimp, macaque, mouse, rat and dog
  • A phylogenetic tree is a tree showing evolutionary history of these species. Leaves are target present-day species, internal nodes are their common ancestors.
  • There are methods to build trees by comparing DNA or protein sequences of several present-day species.
  • Our input contains a small selection of gene sequences from each species. In a real project we would start from all genes (cca 20,000 per species) and would do a careful filtration of problematic sequences, but we skip this step here.
  • The first step will be to identify which genes from different species "correspond" to each other. More exactly, we are looking for groups of orthologs. To do so, we will use a simple method based on sequence similarity, see details below. Again, in real project, more complex methods might be used.
  • The result of ortholog group identification will be a set of genes, each gene having one sequence from each of the 6 species
  • Next we will process each gene separately, aligning them and building a phylogenetic tree for this gene using existing methods.
  • The result of the previous step will be several trees, one for every gene. Ideally, all trees would be identical, showing the real evolutionary history of the six species. But it is not easy to infer the real tree from sequence data, so trees from different genes might differ. Therefore, in the last step, we will build a consensus tree.

Technical overview

This task can be organized in different ways, but to practice Perl, we will write a single Perl script which takes as an input a set of fasta files, each containing DNA sequences of several genes from a single species and writes on output the resulting consensus tree.

  • For most of the steps, we will use existing bioinformatics tools. The script will run these tools and do some additional simple processing.

Temporary directory

  • During its run, the script and various tools will generate many files. All these files will be stored in a single temporary directory which can be then easily deleted by the user.
  • We will use Perl library File::Temp to create this temporary directory with a unique name so that the script can be run several times simultaneously without clashing filenames.
  • The library by default creates the file in /tmp, but instead we will create it in the current directory so that it is not deleted at restart of the computer and so that it can be more easily inspected for any problems
  • The library by default deletes the directory when the script finishes but again, to allow inspection by the user, we will leave the directory in place

Restart

  • The script will have a command line option for restarting the computation and omitting the time-consuming steps that were already finished
  • This is useful in long-running scripts because during development of the script you will want to run it many times as you add more steps. In real usage the computation can also be interrupted for various reasons.
  • Our restart capabilities will be quite rudimentary: before running a potentially slow external program, the script will check if the temporary directory contains a non-empty file with the filename matching the expected output of the program. If the file is found, it is assumed to be correct and complete and the external program is not run.

Command line options

  • The script should be named build-tree.pl and as command-line arguments, it will get names of the species
    • For example, we can run the script as follows: ./build-tree.pl human chimp macaque mouse rat dog
    • The first species, in this case human, will be so called reference species (see task A)
    • The script needs at least 2 species, otherwise it will write an error message and stop
    • For each species X there should be a file X.fa in the current directory, this is also checked by the script
  • Restart is specified by command line option -r followed by the name of temporary directory
  • Command-line option handling and creation of temporary directory is already implemented in the script you are given.

Input files

  • Each input fasta X.fa file contains DNA sequences of several genes from one species X
  • Each sequence name on a line starting with > will contain species name, underscore and gene id, e.g. ">human_00008"
  • Species name matches name of the file, gene id is unique within the fasta file
  • Species names and gene ids do not contain underscore, whitespace or any other special characters
  • Sequence of each gene can be split into several lines

Files and submitting

In /tasks/hw02/ you will find the following files:

  • 6 fasta files (*.fa)
  • skeleton script build-tree.pl
    • This script already contains handling of command line options, entire task B, potentially useful functions my_run and my_delete and suggested function headers for individual tasks. Feel free to change any of this.
  • outline of protocol protocol.txt
  • directory example with files for two different groups of genes

Copy the files to your directory and continue writing the script

Submitting

  • Submit the script, protocol protocol.txt or protocol.pdf and temporary directory with all files created in the run of your script on all 6 species with human as reference.
  • Since the commands and names of files are specified in the homework, you do not need to write them in the protocol (unless you change them). Therefore it is sufficient if the protocol contains self-assessment and any used information sources other than those linked from this assignment or lectures.
  • Submit by copying to /submit/hw02/your_username

Task A: run blast to find similar sequences

  • To find orthologs, we use a simple method by first finding local alignments (regions of sequence similarity) between genes from different species
  • For finding alignments, we will use tool blast (ubuntu package blast2)
  • Example of running blast:
formatdb -p F -i human.fa
blastall -p blastn -m 9 -d human.fa -i mouse.fa -e 1e-5
  • Example of output file:
# BLASTN 2.2.26 [Sep-21-2011]
# Query: mouse_00492
# Database: human.fa
# Fields: Query id, Subject id, % identity, alignment length, mismatches, gap openings, q. start, q. end, s. start, s. end, e-value, bit score
mouse_22930     human_00008     90.79   1107    102     0       1       1107    1       1107    0.0     1386
mouse_22930     human_34035     80.29   350     69      0       745     1094    706     1055    3e-37    147
mouse_22930     human_34035     79.02   143     30      0       427     569     391     533     8e-07   46.1

(note last column - score)

  • For each non-reference species, save the result of blast search in file species.blast in the temporary directory.

Task B: find orthogroups

This part is already implemented in the skeleton file, you don't need to implement or report anything in this task

  • Here, we process all the species.blast files to find ortholog groups.
  • Matches are symmetric, and there can be multiple matches for the same gene. We are looking for reciprocal best hits: pairs of genes human_A and mouse_B, where mouse_B is the match with the highest score in mouse for human_A and human_A is the best-scoring match in human for mouse_B.
  • Some genes in reference species may have no reciprocal best hits in some of the non-reference species.
  • Gene in the reference species and all of its reciprocal best hits constitute orthogroup. If the size of an orthogroup is the same as the number of species, we will call it a complete orthogroup
  • In file genes.txt in temporary directory list we will list all orthogroups, one per line.
chimp_94013 dog_84719 human_15749 macaque_34640 mouse_17461 rat_09232
chimp_61053 human_18570 macaque_12627
chimp_41364 human_19217 macaque_88256 rat_82436

Task C: create a file for each orthogroup

  • For each complete orthogroup, we will create a fasta file with corresponding DNA sequences.
  • The file will be located in temporary directory and will be named genename.fa, where genename is the name of the orthogroup gene from reference species.
  • The fasta name for each sequence is the name of species, NOT the name of the gene.
>human
CTGCGGCTGAGAGAGATGTGTACACTGGGGACGCACTCCGGATCTGCATAGTGACCAAAGAGGGCATCAGGGAGGAAACTGTTTCCTTAAGGAAGGAC
>chimp
TGCGGCTGAGAGAGATGTGTACACTGGGGACGCACTCCGGATCTGCATAGTGACCAAAGAGGGCATCAGGGAGGAGACTGTTTCCTTAAGGAAGGAC
>macaque
CTGCGGCTGAGAGAGACGTGTACACTGGGGACGCGCTCCGGATCTGCATAGTGACCAAAGAGGGCATCAGGGAGGAGACTGTTCCCTTAAGGAAGGAC
>mouse
CAGCCGAGAGGGATGTGTATACTGGAGATGCTCTCAGGATCTGCATCGTGACCAAAGAGGGCATCAGGGAGGAAACTGTTCCCCTGCGGAAAGAC
>rat
CAGCCGAGAGGGATGTGTACACTGGAGACGCCCTCAGGATCTGCATCGTGACCAAAGAGGGCATCAGGGAGGAGACTGTTCCCCTTCGGAAAGAC
>dog
GAGGGATGTGTACACTGGGGATGCACTCAGAATCTGCATTGTGACTAAGGAGGGCATCAGGGAGGAGACTGTTCCCCTGAGGAAGGAT

Task D: build tree for each gene

  • For each orthogroup, we need to build a phylogenetic tree.
  • The result for file genename.fa should be saved in file genename.tree
  • Example of how to do this:
# create multiple alignment of the sequences
muscle -diags -in genename.fa -out genename.mfa
# change format of the multiple alignment
readseq -f12 genename.mfa -o=genename.phy -a
# run phylogenetic inferrence program
phyml -i genename.phy --datatype nt --bootstrap 0 --no_memory_check
# rename the result
mv genename.phy_phyml_tree.txt genename.tree
  • You can view the multiple alignment (*.mfa and *.phy) by using program seaview
  • You can view the resulting tree (*.tree) by using program njplot or figtree

Task E: build consensus tree

  • Trees built on individual genes can differ from each other.
  • Therefore we build a consensus tree: tree that only contains branches present in most gene trees; other branches are collapsed.
  • phylip is an "interactive" program for manipulation of trees. Specific command for building consensus trees is
phylip consense
  • input file for phylip needs to contain all trees of which consensus should be built, one per line
  • text you would type to phylip manually, can be instead passed on the standard input from the script
  • store the output tree from phylip in all_trees.consensus in temporary directory and also print it to standard output

L03

Today: using command-line tools and Perl one-liners.

  • We will do simple transformations of text files using command-line tools without writing any scripts or longer programs.
  • You will record the commands used in your protocol
    • We strongly recommend making a log of commands for data processing also outside of this course
  • If you have a log of executed commands, you can easily execute them again by copy and paste
  • For this reason any comments are best preceded by #
  • If you use some sequence of commands often, you can turn it into a script

Most commands have man pages or are described within man bash

Efficient use of command line

Some tips for bash shell:

  • use tab key to complete command names, path names etc
    • tab completion can be customized [7]
  • use up and down keys to walk through history of recently executed commands, then edit and resubmit chosen command
  • press ctrl-r to search in the history of executed commands
  • at the end of session, history stored in ~/.bash_history
  • command history -a appends history to this file right now
    • you can then look into the file and copy appropriate commands to your protocol
  • various other history tricks, e.g. special variables [8]
  • cd - goes to previously visited directory, also see pushd and popd
  • ls -lt | head shows 10 most recent files, useful for seeing what you have done last

Instead of bash, you can use more advanced command-line environments, e.g. iPhyton notebook

Redirecting and pipes

# redirect standard output to file
command > file

# append to file
command >> file

# redirect standard error
command 2>file

# redirect file to standard input
command < file

# do not forget to quote > in other uses, e.g. when searching for string ">" in a file sequences.fasta
grep '>' sequences.fasta
# (without quotes rewrites sequences.fasta)
# other special characters, such as ;, &, |, # etc should be quoted in '' as well

# send stdout of command1 to stdin of command2
command1 | command2

# backtick operator executes command, 
# removes trailing \n from stdout, substitutes to command line
# the following commands do the same thing:
head -n 2 file
head -n `echo 2` file

# redirect a string in ' ' to stdin of command head
head -n 2 <<< 'line 1
line 2
line 3'

# in some commands, file argument can be taken from stdin if denoted as - or stdin or /dev/stdin
# the following compares uncompressed version of file1 with file2
zcat file1.gz | diff - file2

Make piped commands fail properly:

set -o pipefail

If set, the return value of a pipeline is the value of the last (rightmost) command to exit with a non-zero status, or zero if all commands in the pipeline exit successfully. This option is disabled by default, pipe then returns exit status of the rightmost command.

Text file manipulation

Commands echo and cat (creating and printing files)

# print text Hello and end of line to stdout
echo "Hello" 
# interpret backslash combinations \n, \t etc:
echo -e "first line\nsecond\tline"
# concatenate several files to stdout
cat file1 file2

Commands head and tail (looking at start and end of files)

# print 10 first lines of file (or stdin)
head file
some_command | head 
# print the first 2 lines
head -n 2 file
# print the last 5 lines
tail -n 5 file
# print starting from line 100 (line numbering starts at 1)
tail -n +100 file
# print lines 81..100
head -n 100 file | tail -n 20 

Commands wc, ls -lh, od (exploring file stats and details)

# prints three numbers: number of lines (-l), number of words (-w), number of bytes (-c)
wc file

# prints size of file in human-readable units (K,M,G,T)
ls -lh file

# od -a prints file or stdout with named characters 
#   allows checking whitespace and special characters
echo "hello world!" | od -a
# prints:
# 0000000   h   e   l   l   o  sp   w   o   r   l   d   !  nl
# 0000015

Command grep (getting lines matching a regular expression)

# -i ignores case (upper case and lowercase letters are the same)
grep -i chromosome file
# -c counts the number of matching lines in each file
grep -c '^[12][0-9]' file1 file2

# other options (there is more, see the manual):
# -v print/count not matching lines (inVert)
# -n show also line numbers
# -B 2 -A 1 print 2 lines before each match and 1 line after match
# -E extended regular expressions (allows e.g. |)
# -F no regular expressions, set of fixed strings
# -f patterns in a file 
#    (good for selecting e.g. only lines matching one of "good" ids)

Commands sort, uniq

# some useful options of sort:
# -g numeric sort
# -k which column(s) to use as key
# -r reverse (from largest values)
# -s stable
# -t fields separator

# sorting first by column 2 numerically (-k2,2g), in case of ties use column 1 (-k1,1)
sort -k2,2g -k1,1 file 

# uniq outputs one line from each group of consecutive identical lines
# uniq -c adds the size of each group as the first column
# the following finds all unique lines and sorts them by frequency from the most frequent
sort file | uniq -c | sort -gr

Commands diff, comm (comparing files)

diff compares two files, useful for manual checking of differences

  • useful options
    • -b (ignore whitespace differences)
    • -r for comparing whole directories
    • -q for fast checking for identity
    • -y show differences side-by-side

comm compares two sorted files

  • writes 3 columns:
    • 1: lines occurring only in the first file
    • 2: lines occurring only in the second file
    • 3: lines occurring in both files
  • some columns can be suppressed with -1, -2, -3
  • good for finding set intersections and differences

Commands cut, paste, join (working with columns)

  • cut selects only some columns from file (perl/awk more flexible)
  • paste puts 2 or more files side by side, separated by tabs or other character
  • join is a powerful tool for making joins and left-joins as in databases on specified columns in two files

Commands split, csplit (splitting files to parts)

  • split splits into fixed-size pieces (size in lines, bytes etc.)
  • csplit splits at occurrence of a pattern (e.g. fasta file into individual sequences)
csplit sequences.fa '/^>/' '{*}'

Programs sed and awk

Both programs process text files line by line, allow to do various transformations

  • awk newer, more advanced
  • several examples below
  • More info on wikipedia: awk, sed
# replace text "Chr1" by "Chromosome 1"
sed 's/Chr1/Chromosome 1/'
# prints first two lines, then quits (like head -n 2)
sed 2q  

# print first and second column from a file
awk '{print $1, $2}' 

# print the line if difference in first and second column > 10
awk '{ if ($2-$1>10) print }'  

# print lines matching pattern
awk '/pattern/ { print }' 

# count lines
awk 'END { print NR }'

Perl one-liners

Instead of sed and awk, we will cover Perl one-liners

  • more examples [9], [10]
  • documentation for Perl switches [11]
# -e executes commands
perl -e'print 2+3,"\n"'
perl -e'$x = 2+3; print $x, "\n"';

# -n wraps commands in a loop reading lines from stdin or files listed as arguments
# the following is roughly the same as cat:
perl -ne'print'
# how to use:
perl -ne'print' < input > output
perl -ne'print' input1 input2 > output
# lines are stored in a special variable $_
# this variable is default argument of many functions, 
# including print, so print is the same as print $_

# simple grep-like commands:
perl -ne 'print if /pattern/'
# simple regular expression modifications
perl -ne 's/Chr(\d+)/Chromosome $1/; print'
# // and s/// are applied by default to $_

# -l removes end of line from each input line and adds "\n" after each print
# the following adds * at the end of each line
perl -lne'print $_, "*"' 

# -a splits line into words separated by whitespace and stores them in array @F
# the next example prints difference in numbers stored in the second and first column
# (e.g. interval size if each line coordinates of one interval)
perl -lane'print $F[1]-$F[0]'

# -F allows to set separator used for splitting (regular expression)
# the next example splits at tabs
perl -F '"\t"' -lane'print $F[1]-$F[0]'

# END { commands } is run at the very end, after we finish reading input
# the following example computes the sum of interval lengths
perl -lane'$sum += $F[1]-$F[0]; END { print $sum; }'
# similarly BEGIN { command } before we start

Other interesting possibilites:

# -i replaces each file with a new transformed version (DANGEROUS!)
# the next example removes empty lines from all .txt files in the current directory
perl -lne 'print if length($_)>0' -i *.txt
# the following example replaces sequence of whitespace by exactly one space 
# and removes leading and trailing spaces from lines in all .txt files
perl -lane 'print join(" ", @F)' -i *.txt


# variable $. contains line number. $ARGV name of file or - for stdin
# the following prints filename and line number in front of every line
perl -ne'printf "%s.%d: %s", $ARGV, $., $_' file1 file2

# moving files *.txt to have extension .tsv:
#   first print commands 
#   then execute by hand or replace print with system
#   mv -i asks if something is to be rewritten
ls *.txt | perl -lne '$s=$_; $s=~s/\.txt/.tsv/; print("mv -i $_ $s")'
ls *.txt | perl -lne '$s=$_; $s=~s/\.txt/.tsv/; system("mv -i $_ $s")'

HW03

Lecture 1 (Perl 1), Lecture 2 (Perl 2), Lecture 3 (command-line)

  • In this homework, use command-line tools or one-liners in Perl, awk or sed. Do not write any scripts or programs.
  • Each task can be split into several stages and intermediate files written to disk, but you can also use pipelines to reduce the number of temporary files.
  • Your commands should work also for other input files with the same format (do not try to generalize them too much, but also do not use very specific properties of a particular input, such as the number of lines etc.)
  • Include all relevant used commands in your protocol and add a short description of your approach.
  • Submit the protocol and required output files.
  • Outline of the protocol is in /tasks/hw03/protocol.txt, submit to directory /submit/hw03/yourname


Task A

  • /tasks/hw03/names.txt contains data about several people, one per line.
  • Each line consists of given name(s), surname and email separated by spaces.
  • Each person can have multiple given names (at least 1), but exactly one surname and one email. Email is always of the form username@uniba.sk.
  • The task is to generate file passwords.csv which contains a randomly generated password for each of these users
    • The output file has columns separated by commas ','
    • The first column contains username extracted from email address, the second column surname, the third column all given names and the fourth column the randomly generated password
  • Submit file passwords.csv with the result of your commands.

Example line from input:

Pavol Országh Hviezdoslav hviezdoslav32@uniba.sk

Example line from output (password will differ):

hviezdoslav32,Hviezdoslav,Pavol Országh,3T3Pu3un

Hints:

  • Passwords can be generated using pwgen (e.g. pwgen -N 10 -1 prints 10 passwords, one per line)
  • We also recommend using perl, wc, paste (check option -d in paste)
  • In Perl, function pop may be useful for manipulating @F and function join for connecting strings with a separator.

Task B

File:

  • /tasks/hw03/saccharomyces_cerevisiae.gff contains annotation of the yeast genome
    • Downloaded from http://yeastgenome.org/ on 2016-03-09, in particular from [12].
    • It was further processed to omit DNA sequences from the end of file.
    • The size of the file is 5.6M.
  • For easier work, link the file to your directory by ln -s /tasks/hw03/saccharomyces_cerevisiae.gff yeast.gff
  • The file is in GFF3 format [13]
  • Lines starting with # are comments, other lines contain tab-separated data about one interval of some chromosome in the yeast genome
  • Meaning of the first 5 columns:
    • column 0 chromosome name
    • column 1 source (can be ignored)
    • column 2 type of interval
    • column 3 start of interval (1-based coordinates)
    • column 4 end of interval (1-based coordinates)
  • You can assume that these first 5 columns do not contain whitespace

Task:

  • Print for each type of interval (column 2), how many times it occurs in the file.
  • Sort from the most common to the least common interval types.
  • Hint: commands sort and uniq will be useful. Do not forget to skip comments, for example using grep -v '^#'
  • Submit file types.txt with the output formatted as follows:
   7058 CDS
   6600 mRNA
...
...
      1 telomerase_RNA_gene
      1 mating_type_region
      1 intein_encoding_region

Task C

  • Continue processing file from task B.
  • For each chromosome, the file contains a line which has in column 2 string chromosome, and the interval is the whole chromosome.
  • To file chrosomes.txt, print a tab-separated list of chromosome names and sizes in the same order as in the input
  • The last line of chromosomes.txt should list the total size of all chromosomes combined.
  • Submit file chromosomes.txt
  • Hints:
    • The total size can be computed by a perl one-liner.
    • Example from the lecture: compute the sum of interval sizes if each line of the file contains start and end of one interval: perl -lane'$sum += $F[1]-$F[0]; END { print $sum; }'
    • Grepping for word chromosome does not check if this word is indeed in the second column
    • Tab character is written in Perl as "\t".
  • Your output should start and end as follows:
chrI    230218
chrII   813184
...
...
chrXVI  948066
chrmt   85779
total   12157105

Task D

Overall goal:

  • Proteins from several well-studied yeast species were downloaded from database http://www.uniprot.org/ on 2016-03-09
  • We have also downloaded proteins from yeast Yarrowia lipolytica. We will pretend that nothing is known about these proteins (as if they were produced by gene finding program in a newly sequenced genome).
  • For each Y.lip. proteins we have similar proteins from other yeasts by blast
  • Now we want to find for each protein in Y.lip. its closest match among all known proteins.

Files:

  • /tasks/hw03/known.fa is a fasta file with known proteins from several species
  • /tasks/hw03/yarLip.fa is a fasta file with proteins from Y.lip.
  • /tasks/hw03/known.blast is the result of running blast of yarLip.fa versus known.fa by these commands:
formatdb -i known.fa
blastall -p blastp -d known.fa -i yarLip.fa -m 9 -e 1e-5 > known.blast
  • you can link these files to your directory as follows:
ln -s /tasks/hw03/known.fa .
ln -s /tasks/hw03/yarLip.fa .
ln -s /tasks/hw03/known.blast .

Step 1:

  • Get the first (strongest) match for each query from known.blast.
  • This can be done by printing the lines that are not comments but follow a comment line starting with #.
  • In a perl one-liner, you can create a state variable which will remember if the previous line was a comment and based on that you decide of you print the current line.
  • Instead of using perl, you can play with grep. Option -A 1 prints the matching lines as well as one line ofter each match
  • Print only the first two columns separated by tab (name of query, name of target), sort the file by the second column.
  • Submit file best.tsv with the result
  • File should start as follows:
Q6CBS2  sp|B5BP46|YP52_SCHPO
Q6C8R4  sp|B5BP48|YP54_SCHPO
Q6CG80  sp|B5BP48|YP54_SCHPO
Q6CH56  sp|B5BP48|YP54_SCHPO

Step 2:

  • Submit file known.tsv which contains sequence names extracted from known.fa with leading > removed
  • This file should be sorted alphabetically.
  • File should start as follows:
sp|A0A023PXA5|YA19A_YEAST Putative uncharacterized protein YAL019W-A OS=Saccharomyces cerevisiae (strain ATCC 204508 / S288c) GN=YAL019W-A PE=5 SV=1
sp|A0A023PXB0|YA019_YEAST Putative uncharacterized protein YAR019W-A OS=Saccharomyces cerevisiae (strain ATCC 204508 / S288c) GN=YAR019W-A PE=5 SV=1

Step 3:

  • Use command join to join the files best.tsv and known.tsv so that each line of best.tsv is extended with the text describing the corresponding target in known.tsv
  • Use option -1 2 to use the second column of best.tsv as a key for joining
  • The output of join may look as follows:
sp|B5BP46|YP52_SCHPO Q6CBS2 Putative glutathione S-transferase C1183.02 OS=Schizosaccharomyces pombe (strain 972 / ATCC 24843) GN=SPBC460.02c PE=3 SV=1
sp|B5BP48|YP54_SCHPO Q6C8R4 Putative alpha-ketoglutarate-dependent sulfonate dioxygenase OS=Schizosaccharomyces pombe (strain 972 / ATCC 24843) GN=SPBC460.04c PE=3 SV=1
  • Further reformat the output so that query name goes first (e.g. Q6CBS2), followed by target name (e.g. sp|B5BP46|YP52_SCHPO), followed by the rest of the text, but remove all text after OS=
  • Sort by query name
  • Submit file best.txt with the result
  • The output should start as follows:
B5FVA8  tr|Q5A7D5|Q5A7D5_CANAL  Lysophospholipase
B5FVB0  sp|O74810|UBC1_SCHPO    Ubiquitin-conjugating enzyme E2 1
B5FVB1  sp|O13877|RPAB5_SCHPO   DNA-directed RNA polymerases I, II, and III subunit RPABC5

Note:

  • Not all Y.lip. are necessarily included in your final output (some proteins do not have blast match).
    • You can think how to find the list of such proteins, but this is not part of the assignment.
  • Files best.txt and best.tsv should have the same number of lines.

L04

Job Scheduling

  • Some computing jobs take a lot of time: hours, days, weeks,...
  • We do not want to keep a command-line window open the whole time; therefore we run such jobs in the background
  • Simple commands to do it in Linux:
    • To run the program immediately, then switch the whole console to the background: screen, tmux
    • To run the command when the computer becomes idle: batch
  • Now we will concentrate on Sun Grid Engine, a complex software for managing many jobs from many users on a cluster from multiple computers
  • Basic workflow:
    • Submit a job (command) to a queue
    • The job waits in the queue until resources (memory, CPUs, etc.) become available on some computer
    • The job runs on the computer
    • Output of the job is stored in files
    • User can monitor the status of the job (waiting, running)
  • Complex possibilities for assigning priorities and deadlines to jobs, managing multiple queues etc.
  • Ideally all computers in the cluster share the same environment and filesystem
  • We have a simple training cluster for this exercise:
    • You submit jobs to queue on vyuka
    • They will run on computer cpu02
    • This cluster is only temporarily available until next Thursday

Submitting a job (qsub)

  • qsub -b y -cwd 'command < input > output 2> error'
    • quoting around command allows us to include special characters, such as <, > etc. and not to apply it to qsub command itself
    • -b y treats command as binary, usually preferable for both binary programs and scripts
    • -cwd executes command in the current directory
    • -N name allows to set name of the job
    • -l resource=value requests some non-default resources
    • for example, we can use -l threads=2 to request 2 threads for parallel programs
    • Grid engine will not check if you do not use more CPUs or memory than requested, be considerate (and perhaps occasionally watch your jobs by running top at the computer where they execute)
  • qsub will create files for stdout and stderr, e.g. s2.o27 and s2.e27 for the job with name s2 and jobid 27

Monitoring and deleting jobs (qstat, qdel)

  • qstat displays jobs of the current user
job-ID  prior   name       user         state submit/start at     queue                          slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
     28 0.50000 s3         bbrejova     r     03/15/2016 22:12:18 main.q@cpu02.compbio.fmph.unib     1
     29 0.00000 s3         bbrejova     qw    03/15/2016 22:14:08                                    1
  • qstat -u '*' displays jobs of all users
    • finished jobs disappear from the list
  • qstat -F threads shows how many threads available
queuename                      qtype resv/used/tot. load_avg arch          states
---------------------------------------------------------------------------------
main.q@cpu02.compbio.fmph.unib BIP   0/2/8          0.03     lx26-amd64
        hc:threads=0
     28 0.75000 s3         bbrejova     r     03/15/2016 22:12:18     1
     29 0.25000 s3         bbrejova     r     03/15/2016 22:14:18     1
  • Command qdel allows you to delete a job (waiting or running)

Interactive work on the cluster (qrsh), screen

  • qrsh creates a job which is a normal interactive shell running on the cluster
  • in this shell you can manually run commands
  • when you close the shell, the job finishes
  • therefore it is a good idea to run qrsh within screen
    • run screen command, this creates a new shell
    • within this shell, run qrsh, then whatever commands
    • by pressing Ctrl-a d you "detach" the screen, so that both shells (local and qrsh) continue running but you can close your local window
    • later by running screen -r you get back to your shells

Running many small jobs

For example, consider tens of thousands of genes, run some computation for each gene

  • Have a script which iterates through all and runs them sequentially (as in HW02).
    • Problems: Does not use parallelism, needs more programming to restart after some interruption
  • Submit processing of each gene as a separate job to cluster (submitting done by a script/one-liner).
    • Jobs can run in parallel on many different computers
    • Problem: Queue gets very long, hard to monitor progress, hard to resubmit only unfinished jobs after some failure.
  • Array jobs in qsub (option -t): runs jobs numbered 1,2,3...; number of the job in an environment variable, used by the script to decide which gene to process
    • Queue contains only running sub-jobs plus one line for the remaining part of the array job.
    • After failure, you can resubmit only unfinished portion of the interval (e.g. start from job 173).
  • Next: using make in which you specify how to process each gene and submit a single make command to the queue
    • Make can execute multiple tasks in parallel using several threads on the same computer (qsub array jobs can run tasks on multiple computers)
    • It will automatically skip tasks which are already finished

Make

  • Make is a system for automatically building programs (running compiler, linker etc)
  • Rules for compilation are written in a Makefile
  • Rather complex syntax with many features, we will only cover basics

Rules

  • The main part of a Makefile are rules specifying how to generate target files from some source files (prerequisites).
  • For example the following rule generates target.txt by concatenating source1.txt a source2.txt:
target.txt : source1.txt source2.txt
      cat source1.txt source2.txt > target.txt
  • The first line describes target and prerequisites, starts in the first column
  • The following lines list commands to execute to create the target
  • Each line with a command starts with a tab character
  • If we have a directory with this rule in Makefile and files source1.txt and source2.txt, running make target.txt will run the cat command
  • However, if target.txt already exists, the command will be run only if one of the prerequisites has more recent modification time than the target
  • This allows to restart interrupted computations or rerun necessary parts after modification of some input files
  • Makefile automatically chains the rules as necessary:
    • if we run make target.txt and some prerequisite does not exist, Makefile checks if it can be created by some other rule and runs that rule first
    • In general it first finds all necessary steps and runs them in topological order so that each rules has its prerequisites ready
    • Option make -n target will show what commands would be executed to build target (dry run) - good idea before running something potentially dangerous

Pattern rules

  • We can specify a general rule for files with a systematic naming scheme. For example, to create a .pdf file from a .tex file, we use pdflatex command:
%.pdf : %.tex
      pdflatex $^
  • In the first line, % denotes some variable part of the filename, which has to agree in the target and all prerequisites
  • In commands, we can use several variables:
    • $^ contains name for the prerequisite (source)
    • $@ contains the name of the target
    • $* contains the string matched by %

Other useful tricks in Makefiles

Variables

  • Store some reusable values in variables, then use them several times in the Makefile:
MYPATH := /projects/trees/bin

target : source
       $(MYPATH)/script < $^ > $@

Wildcards, creating a list of targets from files in the directory

The following Makefile automatically creates .png version of each .eps file simply by running make:

EPS := $(wildcard *.eps)
EPSPNG := $(patsubst %.eps,%.png,$(EPS))

all:  $(EPSPNG)

clean:
        rm $(EPSPNG)

%.png : %.eps
        convert -density 250 $^ $@
  • variable EPS contains names of all files matching *.eps
  • variable EPSPNG contains desirable names of png files
    • it is created by taking filenames in EPS and changing .eps to .png
  • all is a "phony target" which is not really created
    • its rule has no commands but all png files are prerequisites, so are done first
    • the first target in Makefile (in this case all) is default when no other target is specified on command-line
  • clean is also a phony target for deleting generated png files

Useful special built-in target names

Include these lines in your Makefile if desired

.SECONDARY:
# prevents deletion of intermediate targets in chained rules

.DELETE_ON_ERROR:
# delete targets if a rule fails

Parallel make

  • running make with option -j 4 will run up to 4 commands in parallel if their dependencies are already finished
  • easy parallelization on a single computer

Alternatives to Makefiles

  • Bioinformatics often uses "pipelines" - sequences of commands run one after another, e.g. by a script of Makefile
  • There are many tools developed for automating computational pipelines, see e.g. this review: Jeremy Leipzig; A review of bioinformatic pipeline frameworks. Brief Bioinform 2016 bbw020.
  • For example Snakemake
    • Workflows can contain shell commands or Python code
    • Big advantage compared to Make: pattern rules may contain multiple variable portions (in make only one % per filename)
    • For example, you have several fasta files and several HMMs representing protein families and you wans to run each HMM on each fasta file:
rule HMMER:
     input: "{filename}.fasta", "{hmm}.hmm"
     output: "{filename}_{hmm}.hmmer"
     shell: "hmmsearch --domE 1e-5 --noali --domtblout {output} {input[1]} {input[0]}"

HW04

See also Lecture 4, Lecture 2, #HW02

In this homework, we will return to the example in homework 2, where we took genes from several organisms, found orthogroups of corresponding genes and built a phylogenetic tree for each orthogroup. This was all done in a single big Perl script. In this homework, we will write a similar pipeline using make and execute it remotely using qsub. We will use proteins instead of DNA and we will use a different set of species. Most of the work is already done, only small modifications are necessary.

  • Submit by copying requested files to /submit/hw04/username/
  • Do not forget to submit protocol, outline of the protocol is in /tasks/hw04/protocol.txt

Task A

  • In this task, you will run a long alignment job (>1 hour)
  • Copy directory /tasks/hw04/large to your home directory
    • ref.fa: all proteins from yeast Yarrowia lipolytica
    • other.fa: all proteins from 8 other yeast species
    • Makefile: run blast on ref.fa vs other.fa (also formats database other.fa before that)
  • run make -n to see what commands will be done (you should see formatdb and blastall + echo for timing), copy the output to the protocol
  • run qsub with appropriate options to run make (at least -cwd and -b y)
  • then run qstat > queue.txt
    • Submit file queue.txt showing your job waiting or running
  • When your job finishes, submit also the following two files:
    • the last 100 lines from the output file ref.blast under the name ref-end.blast (use tool tail -n 100)
    • standard output from the qsub job, which is stored in a file named e.g. make.oX where X is the number of your job. The output shows the time when your job started and finished (this information was written by commands echo in the Makefile)

Task B

  • In this task, you will finish a Makefile for splitting blast results into orthogroups and building phylogenetic trees for each group
    • This Makefile works with much smaller files and so you can run it many times on vyuka, without qsub
    • If it runs too slowly, you can temporarily modify ref.fa to contain only the first 2 sequences, debug your makefile and then again copy the original ref.fa from /tasks/hw04/small to run the final analysis
  • Copy directory /tasks/hw04/small to your home directory
    • ref.fa: 6 proteins from yeast Yarrowia lipolytica
    • other.fa: a selected subset of proteins from 8 other yeast species
    • Makefile: a longer makefile

The Makefile runs the analysis in four stages. Stages 1,2 and 4 are done, you have to finish stage 3

  • If you run make without argument, it will attempt to run all 4 stages, but stage 3 will not run, because it is missing
  • Stage 1: run as make ref.brm
    • It runs blast as in task A, then splits proteins into orthogroups and creates one directory for each group with file prot.fa containing protein sequences
  • Stage 2: run as make alignments
    • In each directory with a single gene, it will create an alignment prot.phy and link it under names lg.phy and wag.phy
  • Stage 3: run as make trees (needs to be written by you)
    • In each directory with a single gene, it should create lg.phy_phyml_tree and wag.phy_phyml_tree
    • These corresponds to results of phyml commands run with two different evolutionary models WAG and LG, where LG is the default
    • Run phyml by commands of the forms:
      • phyml -i INPUT --datatype aa --bootstrap 0 --no_memory_check >LOG
      • phyml -i INPUT --model WAG --datatype aa --bootstrap 0 --no_memory_check >LOG
    • Change INPUT and LOG in the commands to appropriate filenames using make variables $@, $^, $* etc. Input should come from lg.phy or wag.phy in the directory of a gene and log should be the same as tree name with extension .log added (e.g. lg.phy_phyml_tree.log)
    • Also add variables LG_TREES and WAG_TREES listing filenames of all desirable trees and uncomment phony target trees which uses these variables
  • Stage 4: run as make consensus
    • Output trees from stage 3 are concatenated for each model separately to files lg/intree wag/intree and then phylip is run to produce consensus trees lg.tree and wag.tree
    • This stage also needs variables LG_TREES and WAG_TREES to be defined by you.
  • Run your Makefile
  • Submit the whole directory small, including Makefile and all gene directories with tree files.

Task C

  • Look at the two trees from task B (wag.tree, lg.tree) using the figtree program, switch on displaying branch labels in the left panel with options. These labels show for each branch of the tree, how many of the input trees support this branch.
  • Write your observations to the protocol: Do the two trees differ? If yes, do they differ in branches supported by many different genes trees, or few? What is the highest and lowest support for a branch in each tree?
    • Note that the two children of each internal node are equivalent, so their placement higher or lower in the figure does not matter.

Further possibilities

Here are some possibilities for further experiments, in case you are interested (do not submit these):

  • You could copy your extended Makefile to directory large and create trees for all orthogroups in the big set
    • This would take a long time, so submit it through qsub and only some time after the lecture is over to allow classmates to work on task A
    • After ref.brm si done, programs for individual genes can be run in parallel, so you can try running make -j 2 and request 2 threads from qsub
  • Phyml also supports other models, for example JTT (see manual), you could try to play with those.
  • Command touch FILENAME will change modification time of the given file to current file
    • What happens when you run touch on some of the intermediate files in the analysis in task B? Does Makefile always run properly?

L05

#HW05

  • Program for today: basics of Python and SQL, bonus homework for 50% of weight of a regular HW.
  • In the next three lectures (after the Easter), you will use Python and SQLite3 and several advanced Python libraries for complex data processing.

Overview, documentation

Python: good sources for beginners:

  • A very concise cheat sheet: [14]
  • A more detailed tutorial: [15]

SQL:

  • Language for working with relational databases, more in a dedicated course
  • We will cover basics of SQL and work with a simple DB system SQLite3
  • SQLite3 documentation: [16]
  • SQL tutorial: [17]
  • SQLite3 in Python [18]

Program for today:

  • We introduce a simple data set
  • We look at several python scripts for processing this data set
  • HW: You create another such script
  • We introduce basics of working directly with SQLite3
  • HW: You write your own queries
  • We look at how to combine Python and SQLite
  • HW: You write a program combining the two

Dataset for this week

  • IMDb is an online database of movies and TV series with user ratings
  • We have downloaded a preprocessed dataset of selected TV series ratings from GitHub
  • From dataset this we have selected only 10 series with the highest average number of voting users
  • Data are 2 files in csv format: list of series, list of episodes

File series.cvs contains one row per series

  • Columns: (0) series id, (1) series title, (2) TV channel:
3,Breaking Bad,AMC
2,Sherlock,BBC
1,Game of Thrones,HBO 

File episodes.csv contains one row per episode:

  • Columns: (0) series id, (1) episode title, (2) episode order within the whole series, (3) season number, (4) episode number within season, (5) user rating, (6) the number of votes
  • Here is a sample of 4 episodes from Game of Thrones
  • If the episode title contains a comma, the whole tile is in quotation marks
1,"Dark Wings, Dark Words",22,3,2,8.6,12714
1,No One,58,6,8,8.3,20709
1,Battle of the Bastards,59,6,9,9.9,138353
1,The Winds of Winter,60,6,10,9.9,93680

Several python scripts

prog1.py

Print the second column (series tile) from series.csv

#! /usr/bin/python3

# open a file for reading
with open('series.csv') as csvfile:
    # iterate over lines of the input file
    for line in csvfile:
        # split a line into columns at commas
        columns = line.split(",")
        # print the second column
        print(columns[1])

prog2.py

Print list of series of each TV channel

  • For illustration we also separately count the series for each channel, but the count could be obtained as the length of the list
  • For simplicity we use library data structure defaultdict instead of plain python dictionary
#! /usr/bin/python3
from collections import defaultdict

# Create a dictionary in which default value
# for non-existent key is 0 (type int)
# For each channel we willl count the series
channel_counts = defaultdict(int)

# Create a dictionary for keeping a list of series per channel
# default value empty list
channel_lists = defaultdict(list)

# open a file and iterate over lines
with open('series.csv') as csvfile:
    for line in csvfile:
        # strip whitespace (e.g. end of line) from end of line
        line = line.rstrip()
        # split line into columns, find channel and series names
        columns = line.split(",")
        channel = columns[2]
        series = columns[1]
        # increase counter for channel
        channel_counts[channel] += 1
        # add series to list for the channel
        channel_lists[channel].append(series)

# print counts
print("Counts:")
for channel in channel_counts:
    print("The number of series for channel \"%s\" is %d" 
    % (channel, channel_counts[channel]))


# print series lists
print("\nLists:")
for channel in channel_lists:
    list = ", ".join(channel_lists[channel]) 
    print("series for channel \"%s\": %s" % (channel,list))

prog3.py

Find the episode with the highest number of votes among all episodes

  • We use a libary for csv parsing to deal with quotation marks.
#! /usr/bin/python3
import csv

#keep maximum number of votes and its episode
max_votes = 0
max_votes_episode = None

# open a file
with open('episodes.csv') as csvfile:
    # create a reader for parsin csv files
    reader = csv.reader(csvfile, delimiter=',', quotechar='"')
    # iterate over rows already split into columns
    for row in reader:
        votes = int(row[6])
        if votes > max_votes:
            max_votes = votes
            max_votes_episode = row[1]
        
# print result
print("Maximum votes %d in episode \"%s\"" % (max_votes, max_votes_episode))

prog4.py

Example of function definition, reading the whole file into a 2d array

#! /usr/bin/python3
import csv

def read_csv_to_list(filename):
    # create empty list
    rows = []
    # open a file
    with open(filename) as csvfile:
        # create a reader for parsin csv files
        reader = csv.reader(csvfile, delimiter=',', quotechar='"')
        # iterate over rows already split into columns
        for row in reader:
            rows.append(row)
    return rows

series = read_csv_to_list('series.csv')
episodes = read_csv_to_list('episodes.csv')
print("the number of episodes is %d" % len(episodes))
# further processing of series and episodes...

Now do #HW05, task A

SQL and SQLite

Creating a database

SQLite3 database is a file with your data stored in some special format. To load our csv file to a SQLite database, run command:

sqlite3 series.db < create_db.sql

Contents of create_db.pl:

CREATE TABLE series (
  id INT,
  title TEXT,
  channel TEXT
);
.mode csv
.import series.csv series
CREATE TABLE episodes (
  seriesId INT,
  title TEXT,
  orderInSeries INT,
  season INT,
  orderInSeason INT,
  rating REAL,
  votes INT
);
.mode csv
.import episodes.csv episodes

SQL queries

Run sqlite3 series.db

  • the type on SQLite3 command line the following queries
  • The first two only switch on human-friendly formatting
/*  switch on human-friendly formatting */
.mode column
.headers on

/* print title of each series (as prog1.py) */
SELECT title FROM series;

/* sort titles alphabetically */
SELECT title FROM series ORDER BY title;

/* find the highest number among episodes */
SELECT MAX(votes) FROM episodes;

/* find epsiode with the highest number of votes, as prog3.py */
SELECT title, votes FROM episodes
  ORDER BY votes DESC LIMIT 1;

/* print all episodes with at least 50k votes, order by votes */
SELECT title, votes FROM episodes
  WHERE votes>50000 ORDER BY votes desc;

/* join series and episodes tables, print 10 epsiodes
 * with the highest number of votes */
SELECT s.title, e.title, votes
  FROM episodes AS e, series AS s
  WHERE e.seriesId=s.id
  ORDER BY votes desc limit 10;

/* compute the number of series per channel, as prog2.py */
SELECT channel, COUNT() as series_count
  FROM series GROUP BY channel;

/* print the number of episodes and avergae rating per season and series */
SELECT seriesId, season, COUNT() AS episode_count, AVG(rating) AS rating
  FROM episodes GROUP BY seriesId, season;

Now do #HW05, tasks B1, B2

Accessing database from Python

read_db.py

  • Script illustrates running a SELECT query and getting results
#! /usr/bin/python3
import sqlite3

# connect to a database 
connection = sqlite3.connect('series.db')
# create a "cursor" for working with th database
cursor = connection.cursor()

# run a select query
# supply parameters of the query using placeholders ?
threshold = 40000
cursor.execute("""SELECT title, votes FROM episodes
  WHERE votes>? ORDER BY votes desc""", (threshold,))

# retrieve results of the query
for row in cursor:
    print("Episode \"%s\" votes %s" % (row[0],row[1]))
    
# close db connection
connection.close()

write_db.py

Script illustrates creating a new database containing a multiplication table

#! /usr/bin/python3
import sqlite3

# connect to a database 
connection = sqlite3.connect('multiplication.db')
# create a "cursor" for working with th database
cursor = connection.cursor()

cursor.execute("""
CREATE TABLE mult_table (
a INT, b INT, mult INT)
""")

for a in range(1,11):
    for b in range(1,11):
        cursor.execute("INSERT INTO mult_table (a,b,mult) VALUES (?,?,?)",
                       (a,b,a*b))

# important: save the changes
connection.commit()
    
# close db connection
connection.close()

We can check the result by running command

sqlite3 multiplication.db "SELECT * FROM mult_table;"

Now do #HW05, task C

HW05

Lecture 05

Preparation

Copy files:

mkdir hw05
cd hw05
cp -iv /tasks/hw05/* .

The directory contains the following files:

  • *.py: python scripts for the lecture, included only for convenience
  • series.csv, episodes.csv: data file used in the homework (and the lecture)
  • create_db.sql: sql commands to create the database needed in tasks B1, B2, C
  • protocol.txt: fill in and submit the protocol. Only "Vyhodnotenie" and "Pouzite zdroje" are needed this time

To prepare the database for tasks B1, B2 and C, run the command:

sqlite3 series.db < create_db.sql

To verify that your database was created correctly, you can run the following commands:

sqlite3 series.db ".tables"
# output should be  episodes  series  

sqlite3 series.db "select count() from episodes; select count() from series;"
# output should be 348 and 10

Task A

  • Write a script which reads both csv files and outputs for each TV channel the total number of episodes in their series combined
  • Submit file taskA.py with your script
  • Run your script as follows and submit the file taskA.txt:
./taskA.py > taskA.txt
  • One of the lines of your output should be:
The number of episodes for channel "HBO" is 76

Hints:

  • A good place to start is prog4.py with reading both csv files and prog3.py with dictionary of counters
  • It might be useful to build a dictionary linking series id to the channel name for that series

Task B1

  • Prepare your database as shown above
  • The last query in the lecture counts the number of episodes and average rating per each season of each series
SELECT seriesId, season, COUNT() AS episode_count, AVG(rating) AS rating
  FROM episodes GROUP BY seriesId, season;
  • Use join with series table to replace numeric series id with series title and add the channel name
  • Write your SQL query to file taskB1.sql and submit this file
    • The first two lines of the sql file should be
.mode column
.headers on
  • Run your query as follows:
sqlite3 series.db < taskB1.sql > taskB1.txt
  • Submit also the resulting file taskB1.txt
  • For example, both seasons of True Detective by HBO have 8 episodes and average ratings 9.3 and 8.25
True Detective   HBO         1           8              9.3       
True Detective   HBO         2           8              8.25      

Task B2

  • For each channel compute the total count and average rating of all their episodes.
  • Write your SQL query to file taskB2.sql and submit this file
    • The first two lines of the sql file should be
.mode column
.headers on
  • Run your query as follows:
sqlite3 series.db taskB2.sql > taskB2.txt
  • Submit also the resulting file taskB2.txt
  • For example, all 76 episodes for the two HBO series have average rating as follows:
HBO         76          8.98947368421053

Task C

  • Write a python script that runs the last query from the lecture (shown below) and stores its results in a separate table called seasons in the series.db database
/* print the number of episodes and average rating per season and series */
SELECT seriesId, season, COUNT() AS episode_count, AVG(rating) AS rating
  FROM episodes GROUP BY seriesId, season;
  • SQL can store results from a query directly in a table, but in this task you should instead read each row of the SELECT query in python and to store it by running INSERT command from python
  • Also do not forget to create the new table in the database with appropriate column names and types. You can execute CREATE TABLE command from python
  • The cursor from the SELECT query is needed while you iterate over the results. Therefore create two cursors - one for reading the database and one for writing.
  • If you change you database during debugging, you can start over by running the command for creating the database above
  • Store and submit the script in taskC.py. Also submit the modified database series.db

Further possibilities

  • If you want to practice Python and SQL some more, you can try this task. Do not submit it.
  • Find all series in which there was a drop in ratings from one season to the next more than 0.5
    • For example in task B1, we have seen drop of 9.3-8.25=1.05 in the True Detective series
  • Analogously you could find series with big increases in the successive seasons
  • One option is to run a query in SQL in which you join table seasons from task C with itself and select rows that belong to the same series and successive seasons
  • Another option is to iterate over all rows of seasons table in Python and to find the answer by comparing rows for successive seasons of the same series

L06

#HW06

In this lecture we dive into SQLite3 and Python.

SQLite3

SQLite3 is a simple "database" stored in one file. Think of SQLite not as a replacement for Oracle but as a replacement for fopen(). Documentation: https://www.sqlite.org/docs.html

You can access sqlite database either from command line:

usamec@Darth-Labacus-2:~$ sqlite3 db.sqlite3
SQLite version 3.8.2 2013-12-06 14:53:30
Enter ".help" for instructions
Enter SQL statements terminated with a ";"
sqlite> CREATE TABLE test(id integer primary key, name text);
sqlite> .schema test
CREATE TABLE test(id integer primary key, name text);
sqlite> .exit

Or from python interface: https://docs.python.org/2/library/sqlite3.html.

Python

Python is a perfect language for almost anything. Here is a cheatsheet: http://www.cogsci.rpi.edu/~destem/igd/python_cheat_sheet.pdf

Scraping webpages

The simplest tool for scraping webpages is urllib2: https://docs.python.org/2/library/urllib2.html Example usage:

import urllib2
f = urllib2.urlopen('http://www.python.org/')
print f.read()

Or use requests package:

import requests
r = requests.get("http://en.wikipedia.org")
print(r.text[:10])

Parsing webpages

We use beautifulsoup4 for parsing html (http://www.crummy.com/software/BeautifulSoup/bs4/doc/). I recommend following examples at the beginning of the documentation and example about CSS selectors: http://www.crummy.com/software/BeautifulSoup/bs4/doc/#css-selectors

Parsing dates

You have two options. Either use datetime.strptime or use dateutil package.

Other usefull tips

  • Don't forget to commit to your sqlite3 database (db.commit()).
  • CREATE TABLE IF NOT EXISTS can be usefull at the start of your script.
  • Inspect element (right click on element) in Chrome can be very helpful.
  • Use screen command for long running scripts.
  • All packages are installed on vyuka server. If you are planning using your own laptop, you need to install them using pip (preferably using virtualenv).

HW06

Lecture 6

  • Submit by copying requested files to /submit/hw06/username/

General goal: Scrape comments from several (hundreds) sme.sk users from last month and store them in SQLite3 database.

Task A

Create SQLite3 "database" with appropriate schema for storing comments from SME.sk discussions. You will probably need tables for users and comments. You don't need to store which comments replies to which one.

Submit two files:

  • db.sqlite3 - the database
  • schema.txt - brief description of your schema and rationale behind it

Task B

Build a crawler, which crawls comments in sme.sk discussions. You have two options:

  • For fewer points: Script which gets url of the user (http://ekonomika.sme.sk/diskusie/user_profile.php?id_user=157432) and crawls his comments from last month.
  • For more points: Scripts which gets one starting url (either user profile or some discussion, your choice) and automatically discovers users and crawls their comments.

This crawler should store comments in SQLite3 database built in previous task. Submit following:

  • db.sqlite3 - the database
  • every python script used for crawling
  • README (how to start your crawler)

L07

#HW07

In this lecture we will use Flask and simple text processing utilities from ScikitLearn.

Flask

Flask is simple web server for python (http://flask.pocoo.org/docs/0.10/quickstart/#a-minimal-application) You can find sample flask application at /tasks/hw07/simple_flask. Before running change the port number. You can then access your app at vyuka.compbio.fmph.uniba.sk:4247 (change port number).

There may be problem with access to strange port numbers due to firewalling rules. There are at least two ways to circumvent this:

  • Use X forwarding and run web browser directly from vyuka
local_machine> ssh vyuka.compbio.fmph.uniba.sk -XC
vyuka> chromium-browser
  • Create SOCKS proxy to vyuka.compbio.fmph.uniba.sk and set SOCKS proxy at that port on your local machine. Then all web traffic goes through vyuka.compbio.fmph.uniba.sk via ssh tunnel. To create SOCKS proxy server on local machine port 8000 to vyuka.compbio.fmph.uniba.sk:
local_machine> ssh vyuka.compbio.fmph.uniba.sk -D 8000

(keep ssh session open while working)

Flask uses jinja2 (http://jinja.pocoo.org/docs/dev/templates/) templating language for showing html (you can use strings in python but it is painful).

Processing text

Main tool for processing text is CountVectorizer class from ScikitLearn (http://scikit--learn.org/stable/modules/generated/sklearn.feature_extraction.text.CountVectorizer.html). It transforms text into bag of words (for each word we get counts). Example:

from sklearn.feature_extraction.text import CountVectorizer

vec = CountVectorizer(strip_accents='unicode')

texts = [
 "Ema ma mamu.",
 "Zirafa sa vo vani kupe a hneva sa."
]

t = vec.fit_transform(texts).todense()

print(t)

print(vec.vocabulary)

Useful things

We are working with numpy arrays here (that's array t in example above) Numpy arrays has also lots of nice tricks. First lets create two matrices:

>>> import numpy as np
>>> a = np.array([[1,2,3],[4,5,6]])
>>> b = np.array([[7,8],[9,10],[11,12]])
>>> a
array([[1, 2, 3],
       [4, 5, 6]])
>>> b
array([[7,  8],
       [ 9, 10],
       [11, 12]])

We can sum this matrices or multiply them by some number:

>>> 3 * a
array([[3,  6,  9],
       [12, 15, 18]])
>>> a + 3 * a
array([[4,  8, 12],
       [16, 20, 24]])

We can calculate sum of elements in each matrix, or sum by some axis:

>>> np.sum(a)
21
>>> np.sum(a, axis=1)
array([ 6, 15])
>>> np.sum(a, axis=0)
array([5, 7, 9])

There is a lot other useful functions check https://docs.scipy.org/doc/numpy-dev/user/quickstart.html.

This can help you get top words for each user: http://docs.scipy.org/doc/numpy/reference/generated/numpy.argsort.html#numpy.argsort

HW07

Lecture 7

  • Submit by copying requested files to /submit/hw07/username/

General goal: Build a simple website, which lists all crawled users and for each users has a page with simple statistics for given user.

This lesson requires crawled data from previous lesson, if you don't have one, you can find it at (and thank Baska): /tasks/hw07/db.sqlite3

Submit source code (web server and preprocessing scripts) and database files.

Task A

Create a simple flask web application which:

  • Has a homepage where is a list of all users (with links to their pages).
  • Has a page for each user, which has simple information about user: His nickname, number of posts and hist last 10 posts.

Task B

For each user preprocess and store list of his top 10 words and list of top 10 words typical for him (which he uses much more often than other users, come up with some simple heuristics). Show this information on his page.

Task C

Preprocess and store list of top three similar users for each user (try to come up with some simple definition of similarity based on text in posts). Again show this information on user page.

Bonus: Try to use some simple topic modeling (e.g. PCA as in TruncatedSVD from scikit-learn) and use it for finding similar users.

L08

#HW08

In this lesson we make simple javascript visualizations.

Your goal is to take examples from here https://developers.google.com/chart/interactive/docs/ and tweak them for your purposes.

Tips:

  • You can output your data into javascript data structures in Flask template. It is a bad practice, but sufficient for this lesson. (Better way is to load JSON through API).
  • Remember that you have to bypass the firewall.

HW08

#L08

  • Submit by copying requested files to /submit/hw08/username/

General goal: Extend user pages from previous project with simple visualizations.

Task A

Show a calendar, which shows during which days was user active (like this https://developers.google.com/chart/interactive/docs/gallery/calendar#overview).

Task B

Show a histogram of comments length (like this https://developers.google.com/chart/interactive/docs/gallery/histogram#example).

Task C

Try showing a word tree for a user (https://developers.google.com/chart/interactive/docs/gallery/wordtree#overview). Try to normalize the text (lowercase, remove accents). CountVectorizer has method build_analyzer, which returns a function, which does this for you.

L09

#HW09

Program for today: basics of R (applied to biology examples)

  • very short intro as a lecture
  • tutorial as HW: read a bit of text, try some commands, extend/modify them as requested

In this course we cover several languages popular for scripting in bioinformatics: Perl, Python, R

  • their capabilities overlap, many extensions emulate strengths of one in another
  • choose a language based on your preference, level of knowledge, existing code for the task, rest of the team
  • quickly learn a new language if needed
  • also possibly combine, e.g. preprocess data in Perl or Python, then run statistical analyses in R, automate entire pipeline with bash or make

Introduction

  • R is an open-source system for statistical computing and data visualization
  • Programming language, command-line interface
  • Many built-in functions, additional libraries
  • We will concentrate on useful commands rather than language features

Working in R

  • Run command R, type commands in command-line interface
    • supports history of commands (arrows, up and down, Ctrl-R) and completing command names with tab key
> 1+2
[1] 3
  • Write a script to file, run it from command-line: R --vanilla --slave < file.R
  • Use rstudio to open a graphics IDE [19]
    • Windows with editor of R scripts, console, variables, plots
    • Ctrl-Enter in editor executes current command in console
x=c(1:10)
plot(x,x*x)
  • ? plot displays help for plot command

Suggested workflow

  • work interactively in Rstudio or on command line, try various options
  • select useful commands, store in a script
  • run script automatically on new data/new versions, potentially as a part of a bigger pipeline

Additional information

Gene expression data

  • Gene expression: DNA->mRNA->protein
  • Level of gene expression: Extract mRNA from a cell, measure amounts of mRNA
  • Technologies: microarray, RNA-seq

Gene expression data

  • Rows: genes
  • Columns: experiments (e.g. different conditions or different individuals)
  • Each value is expression of a gene, i.e. relative amount of mRNA for this gene in the sample

We will use microarray data for yeast:

  • Strassburg, Katrin, et al. "Dynamic transcriptional and metabolic responses in yeast adapting to temperature stress." Omics: a journal of integrative biology 14.3 (2010): 249-259. [20]
  • Downloaded from GEO database [21]
  • Data already preprocessed: normalization, log2, etc
  • We have selected only cold conditions, genes with absolute change at least 1
  • Data: 2738 genes, 8 experiments in a time series, yeast moved from normal temperature 28 degrees C to cold conditions 10 degrees C, samples taken after 0min, 15min, 30min, 1h, 2h, 4h, 8h, 24h in cold

HW09

#L09

In this homework, try to read text, execute given commands, potentially trying some small modifications.

  • Then do tasks A-D, submit required files (3x .png)
  • In your protocol, enter commands used in tasks A-D, with explanatory comments in more complicated situations
  • In task B also enter required output to protocol

First steps

  • Type a command, R writes the answer, e.g.:
> 1+2
[1] 3
  • We can store values in variables and use them later on
> # The size of the sequenced portion of cow's genome, in millions of base pairs
> Cow_genome_size <- 2290
> Cow_genome_size
[1] 2290
> Cow_chromosome_pairs <- 30
> Cow_avg_chrom <- Cow_genome_size / Cow_chromosome_pairs
> Cow_avg_chrom
[1] 76.33333

Surprises:

  • dots are used as parts of id's, e.g. read.table is name of a single function (not method for object read)
  • assignment via <- or =
    • careful: a<-3 is an assignment, a < -3 is a comparison
  • vectors etc are indexed from 1, not from 0

Vectors, basic plots

  • Vector is a sequence of values of the same type (all are numbers or all are strings or all are booleans)
# Vector can be created from a list of numbers by function c
a<-c(1,2,4)
a
# prints [1] 1 2 4

# function c also concatenates vectors
c(a,a)
# prints [1] 1 2 4 1 2 4

# Vector of two strings 
b<-c("hello", "world")

# Create a vector of numbers 1..10
x<-1:10
x
# prints [1]  1  2  3  4  5  6  7  8  9 10

Vector arithmetics

  • Operations applied to each member of the vector
x<-1:10
# Square each number in vector x
x*x
# prints [1]   1   4   9  16  25  36  49  64  81 100

# New vector y: logarithm of a number in x squared
y<-log(x*x)
y
# prints [1] 0.000000 1.386294 2.197225 2.772589 3.218876 3.583519 3.891820 4.158883
# [9] 4.394449 4.605170

# Draw graph of function log(x*x) for x=1..10
plot(x,y)
# The same graph but use lines instead of dots
plot(x,y,type="l")

# Addressing elements of a vector: positions start at 1
# Second element of the vector 
y[2]
# prints [1] 1.386294

# Which elements of the vector satisfy certain condition? (vector of logical values)
y>3
# prints [1] FALSE FALSE FALSE FALSE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE

# write only those elements from y that satisfy the condition
y[y>3]
# prints [1] 3.218876 3.583519 3.891820 4.158883 4.394449 4.605170

# we can also write values of x such that values of y satisfy the condition...
x[y>3]
# prints [1]  5  6  7  8  9 10

Task A

  • Create a plot of the binary logarithm with dots in the graph more densely spaced (from 0.1 to 10 with step 0.1)
  • Store it in file log.png and submit this file
  • Hints:
    • Create x and y by vector arithmetics
    • To compute binary logarithm check help ? log
    • Before running plot, use command png("log.png") to store the result, afterwards call dev.off() to close the file (in rstudio you can also export plots manually)

Data frames and simple statistics

  • Data frame: a table similar to spreadsheet, each column is a vector, all are of the same length
  • We will use a table with the following columns:
    • The size of a genome, in millions of nucleotides
    • Number of chromosome pairs
    • GC content
    • Taxonomic group mammal or fish
  • Stored in CSV format, columns separated by tabs.
  • Data: Han et al Genome Biology 2008 [22]
Species    Size    Chrom   GC      Group
Human      2850    23      40.9    mammal
Chimpanzee 2750    24      40.7    mammal
Macaque    2650    21      40.7    mammal
Mouse      2480    20      41.7    mammal
...
Tetraodon   187    21      45.9    fish
...
# reading a frame from file
a<-read.table("/tasks/hw09/genomes.csv", header = TRUE, sep = "\t");
# column with name size
a$Size

# Average chromosome length: divide size by the number of chromosomes
a$Size/a$Chrom

# Add average chromosome length as a new column to frame a
a<-cbind(a,AvgChrom=a$Size/a$Chrom)

# Scatter plot of average chromosome length vs GC content
plot(a$AvgChrom, a$GC)

# Compactly display structure of a 
# (good for checking that import worked etc)
str(a)

# display mean, median, etc. of each column
summary(a);

# average genome size
mean(a$Size)
# average genome size for mammals
mean(a$Size[a$Group=="mammal"])
# Standard deviation
sd(a$Size)

# Histogram of genome sizes
hist(a$Size)

Task B

  • Divide frame a to two frames, one for mammals, one for fish. Hint:
    • Try command a[c(1,2,3),]. What is it doing?
    • Try command a$Group=="mammal".
    • Combine these two commands to get rows for all mammals and store the frame in a new variable, then repeat for fish
    • Use a general approach which does not depend on the exact number and ordering of rows in the table.
  • Run the command summary separately for mammals and for fish. Which of their characteristics are different?
    • Write output and your conclusion to the protocol

Task C

  • Draw a graph comparing genome size vs GC content; use different colors for points representing mammals and fish
    • Submit the plot in file genomes.png
    • To draw the graph, you can use one of the options below, or find yet another way
    • Option 1: first draw mammals with one color, then add fish in another color
      • Color of points can be changed by: plot(1:10,1:10, col="red")
      • After plot command you can add more points to the same graph by command points, which can be used similarly as plot
      • Warning: command points does not change the ranges of x and y axes. You have to set these manually so that points from both groups are visible. You can do this using options xlim and ylim, e.g. plot(x,y, col="red", xlim=c(1,100), ylim=c(1,100))
    • Option 2: plot both mammals and fish in one plot command, and give it a vector of colors, one for each point
      • plot(1:10,1:10,col=c(rep("red",5),rep("blue",5))) will plot the first 5 points red and the last 5 points blue
  • Bonus task: add a legend to the plot, showing which color is mammal and which is fish

Expression data and clustering

Data here is bigger, better to use plain R rather than rstudio (limited server CPU/memory)

# Read gene expression data table
a<-read.table("/tasks/hw09/microarray.csv", header = TRUE, sep = "\t", row.names=1)
# Visual check of the first row
a[1,]
# plot starting point vs. situation after 1 hour
plot(a$cold_0min,a$cold_1h)
# to better see density in dense clouds of points, use this plot
smoothScatter(a$cold_15min,a$cold_1h)
# outliers away from diagonal in the plot above are most strongly differentially expressed genes
# these are more easy to see in MA plot:
# x-axis: average expression in the two conditions
# y-axis: difference between values (they are log-scale, so difference 1 means 2-fold)
smoothScatter((a$cold_15min+a$cold_1h)/2,a$cold_15min-a$cold_1h)

Clustering is a wide group of methods that split data points into groups with similar properties

  • We will group together genes that have a similar reaction to cold, i.e. their rows in gene expression data matrix have similar values

We will consider two simple clustering methods

  • K means clustering splits points (genes) into k clusters, where k is a parameter given by the user. It finds a center of each cluster and tries to minimize the sum of distances from individual points to the center of their cluster. Note that this algorithm is randomized so you will get different clusters each time.
  • Hierarchical clustering puts all data points (genes) to a hierarchy so that smallest subtrees of the hierarchy are the most closely related groups of points and these are connected to bigger and more loosely related groups.
Example of a heatmap
# Heatmap: creates hierarchical clustering of rows 
# then shows every value in the table using color ranging from red (lowest) to white (highest)
# Computation may take some time
heatmap(as.matrix(a),Colv=NA)
# Previous heatmap normalized each row, the next one uses data as they are:
heatmap(as.matrix(a),Colv=NA,scale="none")
# k means clustering to 7 clusters
k=7
cl <- kmeans(a,k)
# each gene has assigned a cluster (number between 1 and k)
cl$cluster
# draw only cluster number 3 out of k
heatmap(as.matrix(a[cl$cluster==3,]),Rowv=NA, Colv=NA)

# reorder genes in the table according to cluster
heatmap(as.matrix(a[order(cl$cluster),]),Rowv=NA, Colv=NA)

# compare overall column means with column means in cluster 3
# function apply uses mean on every column (or row if 2 changed to 1)
apply(a,2,mean)
# now means within cluster
apply(a[cl$cluster==3,],2,mean)

# clusters have centers which are also computed as means
# so this is the same as previous command
cl$centers[3,]

Task D

Example of a required plot
  • Draw a plot in which x-axis is time and y-axis is the expression level and the center of each cluster is shown as a line
    • use command matplot(x,y,type="l") which gets two matrices x and y and plots columns of x vs columns of y
    • matplot(,y,type="l") will use numbers 1,2,3... as columns of the missing matrix x
    • create y from cl$centers by applying function t (transpose)
    • to create an appropriate matrix x, create a vector of times for individual experiments in minutes or hours (do it manually, no need to parse column names automatically)
    • using functions rep and matrix you can create a matrix x in which this vector is used as every column
    • then run matplot(x,y,type="l")
    • since time points are not evenly spaced, it would be better to use logscale: matplot(x,y,type="l",log="x")
    • to avoid log(0), change the first timepoint from 0min to 1min
  • Submit file clusters.png with your final plot

L10

#HW10

Topic of this lecture are statistical tests in R.

  • Beginners in statistics: listen to lecture, then do tasks A, B, C
  • If you know basics of statistical tests, do tasks B, C, D
  • More information on this topic in 1-EFM-340 Počítačová štatistika

Introduction to statistical tests: sign test

  • [23]
  • Two friends A and B have played their favourite game n=10 times, A has won 6 times and B has won 4 times.
  • A claims that he is a better player, B claims that such a result could easily happen by chance if they were equally good players.
  • Hypothesis of player B is called null hypothesis that the pattern we see (A won more often than B) is simply a result of chance
  • Null hypothesis reformulated: we toss coin n times and compute value X: the number of times we see head. The tosses are independent and each toss has equal probability of being 0 or 1
  • Similar situation: comparing programs A and B on several inputs, counting how many times is program A better than B.
# simulation in R: generate 10 psedorandom bits
# (1=player A won)
sample(c(0,1), 10, replace = TRUE)
# result e.g. 0 0 0 0 1 0 1 1 0 0

# directly compute random variable X, i.e. sum of bits
sum(sample(c(0,1), 10, replace = TRUE))
# result e.g. 5

# we define a function which will m times repeat 
# the coin tossing experiment with n tosses 
# and returns a vector with m values of random variable X
experiment <- function(m, n) {
  x = rep(0, m)     # create vector with m zeroes
  for(i in 1:m) {   # for loop through m experiments
    x[i] = sum(sample(c(0,1), n, replace = TRUE)) 
  }
  return(x)         # return array of values     
}
# call the function for m=20 experiments, each with n tosses
experiment(20,10)
# result e.g.  4 5 3 6 2 3 5 5 3 4 5 5 6 6 6 5 6 6 6 4
# draw histograms for 20 experiments and 1000 experiments
png("hist10.png")  # open png file
par(mfrow=c(2,1))  # matrix of plots with 2 rows and 1 column
hist(experiment(20,10))
hist(experiment(1000,10))
dev.off() # finish writing to file
  • It is easy to realize that we get binomial distribution (binomické rozdelenie)
  • <math>Pr(X=k) = {n \choose k} 2^{-n}</math>
  • P-value of the test is the probability that simply by chance we would get k the same or more extreme than in our data.
  • In other words, what is the probability that in 10 tosses we see head 6 times or more (one sided test)
  • <math>\sum_{j=k}^n {n \choose k} 2^{-n}</math>
  • If the p-value is very small, say smaller than 0.01, we reject the null hypothesis and assume that player A is in fact better than B
# computing the probability that we get exactly 6 heads in 10 tosses
dbinom(6, 10, 0.5) # result 0.2050781
# we get the same as our formula above:
7*8*9*10/(2*3*4*(2^10)) # result 0.2050781

# entire probability distribution: probabilities 0..10 heads in 10 tosses
dbinom(0:10, 10, 0.5)
# [1] 0.0009765625 0.0097656250 0.0439453125 0.1171875000 0.2050781250
# [6] 0.2460937500 0.2050781250 0.1171875000 0.0439453125 0.0097656250
# [11] 0.0009765625

#we can also plot the distribution
plot(0:10, dbinom(0:10, 10, 0.5))
barplot(dbinom(0:10,10,0.5))

#our p-value is sum for 7,8,9,10
sum(dbinom(6:10,10,0.5))
# result: 0.3769531
# so results this "extreme" are not rare by chance,
# they happen in about 38% of cases

# R can compute the sum for us using pbinom 
# this considers all values greater than 5
pbinom(5, 10, 0.5, lower.tail=FALSE)
# result again 0.3769531

# if probability is too small, use log of it
pbinom(9999, 10000, 0.5, lower.tail=FALSE, log.p = TRUE)
# [1] -6931.472
# the probability of getting 10000x head is exp(-6931.472) = 2^{-100000}

# generating numbers from binomial distribution 
# - similarly to our function experiment
rbinom(20, 10, 0.5)
# [1] 4 4 8 2 6 6 3 5 5 5 5 6 6 2 7 6 4 6 6 5

# running the test
binom.test(6, 10, p = 0.5, alternative="greater")
#
#        Exact binomial test
#
# data:  6 and 10
# number of successes = 6, number of trials = 10, p-value = 0.377
# alternative hypothesis: true probability of success is greater than 0.5
# 95 percent confidence interval:
# 0.3035372 1.0000000
# sample estimates:
# probability of success
#                   0.6

# to only get p-value run
binom.test(6, 10, p = 0.5, alternative="greater")$p.value
# result 0.3769531

Comparing two sets of values: Welch's t-test

  • Let us now consider two sets of values drawn from two normal distributions with unknown means and variances
  • The null hypothesis of the Welch's t-test is that the two distributions have equal means
  • The test computes test statistics (in R for vectors x1, x2):
    • (mean(x1)-mean(x2))/sqrt(var(x1)/length(x1)+var(x2)/length(x2))
  • This test statistics is approximately distributed according to Student's t-distribution with the degree of freedom obtained by
n1=length(x1)
n2=length(x2)
(var(x1)/n1+var(x2)/n2)**2/(var(x1)**2/((n1-1)*n1*n1)+var(x2)**2/((n2-1)*n2*n2))
  • Luckily R will compute the test for us simply by calling t.test
x1 = rnorm(6, 2, 1)
# 2.70110750  3.45304366 -0.02696629  2.86020145  2.37496993  2.27073550

x2 = rnorm(4, 3, 0.5)
# 3.258643 3.731206 2.868478 2.239788
> t.test(x1,x2)
# t = -1.2898, df = 7.774, p-value = 0.2341
# alternative hypothesis: true difference in means is not equal to 0
# means 2.272182  3.024529

x2 = rnorm(4, 5, 0.5)
# 4.882395 4.423485 4.646700 4.515626
t.test(x1,x2)
# t = -4.684, df = 5.405, p-value = 0.004435
# means 2.272182  4.617051

# to get only p-value, run 
t.test(x1,x2)$p.value

We will apply Welch's t-test to microarray data

  • Data from GEO database [24], publication [25]
  • Abbott et al 2007: Generic and specific transcriptional responses to different weak organic acids in anaerobic chemostat cultures of Saccharomyces cerevisiae
  • gene expression measurements under 5 conditions:
    • reference: yeast grown in normal environment
    • 4 different acids added so that cells grow 50% slower (acetic, propionic, sorbic, benzoic)
  • from each condition (reference and each acid) we have 3 replicates
  • together our table has 15 columns (3 replicates from 5 conditions)
  • 6398 rows (genes)
  • We will test statistical difference between the reference condition and one of the acids (3 numbers vs other 2 numbers)
  • See Task B in #HW10

Multiple testing correction

  • When we run t-tests on the reference vs. acetic acid on all 6398 genes, we get 118 genes with p-value<=0.01
  • Purely by chance this would happen in 1% of cases (from definition of p-value)
  • So purely by chance we would expect to get about 64 genes with p-value<=0.01
  • So perhaps roughly half of our detected genes (maybe less, maybe more) are false positives
  • Sometimes false positives may even overwhelm results
  • Multiple testing correction tries to limit the number of false positives among results of multiple statistical tests
  • Many different methods
  • The simplest one is Bonferroni correction, where the threshold on p-value is divided by the number of tested genes, so instead of 0.01 we use 0.01/6398 = 1.56e-6
  • This way the expected overall number of false positives in the whole set is 0.01 and so the probability of getting even a single false positive is also at most 0.01 (by Markov inequality)
  • We could instead multiply all p-values by the number of tests and apply the original threshold 0.01 - such artificially modified p-values are called corrected
  • After Bonferroni correction we get only 1 significant gene
# the results of p-tests are in vector pa of length 6398
# manually multiply p-values by length(pa), count those that have value <=0.01
sum(pa * length(pa) < 0.01)
# in R you can use p.adjust form multiple testing correction
pa.adjusted = p.adjust(pa, method ="bonferroni")
# this is equivalent to multiplying by the length and using 1 if the result > 1
pa.adjusted = pmin(pa*length(pa),rep(1,length(pa)))

# there are less conservative multiple testing correction methods, e.g. Holm's method
# but in this case we get almost the same results
pa.adjusted2 = p.adjust(pa, method ="holm")
  • Other frequently used correction is false discovery rate (FDR), which is less strict and controls the overall proportion of false positives among results

HW10

#L10

  • Do either tasks A,B,C (beginners) or B,C,D (more advanced). If you really want, you can do all four for bonus credit.
  • In your protocol write used R commands with brief comments on your approach.
  • Submit required plots with filenames as specified.
  • For each task also include results as required and a short discussion commenting the results/plots you have obtained. Is the value of interest increasing or decreasing with some parameter? Are the results as expected or surprising?
  • Outline of protocol is in /tasks/hw10/protocol.txt

Task A: sign test

  • Consider a situation in which players played n games, out of which a fraction of q were won by A (example in lecture corresponds to q=0.6 and n=10)
  • Compute a table of p-values for n=10,20,...,90,100 and for q=0.6, 0.7, 0.8, 0.9
  • Plot the table using matplot (n is x-axis, one line for each value of q)
  • Submit the plot in sign.png
  • Discuss the values you have seen in the plot / table

Outline of the code:

# create vector rows with values 10,20,...,100
rows=(1:10)*10
# create vector columns with required values of q
columns=c(0.6, 0.7, 0.8, 0.9)
# create empty matrix of pvalues 
pvalues = matrix(0,length(rows),length(columns))
# TODO: fill in matrix pvalues using binom.test

# set names of rows and columns
rownames(pvalues)=rows
colnames(pvalues)=columns
# careful: pvalues[10,] is now 10th row, i.e. value for n=100, 
#          pvalues["10",] is the first row, i.e. value for n=10

# check that for n=10 and q=0.6 you get p-value 0.3769531
pvalues["10","0.6"]

# create x-axis matrix (as in HW09, part D)
x=matrix(rep(rows,length(columns)),nrow=length(rows))
# matplot command
png("sign.png")
matplot(x,pvalues,type="l",col=c(1:length(columns)),lty=1)
legend("topright",legend=columns,col=c(1:length(columns)),lty=1)
dev.off()

Task B: Welch's t-test on microarray data

  • Read table with microarray data, transform it to log scale, then work with table a:
input=read.table("/tasks/hw10/acids.tsv", header=TRUE, row.names=1)
a = log(input)
  • Columns 1,2,3 are reference, columns 4,5,6 acetic acid, 7,8,9 benzoate, 10,11,12 propionate, and 13,14,15 sorbate
  • Write a function my.test which will take as arguments table a and 2 lists of columns (e.g. 1:3 and 4:6) and will run for each row of the table Welch's t-test of the first set of columns vs the second set. It will return the resulting vector of p-values
  • For example by calling pa <- my.test(a, 1:3, 4:6) we will compute p-values for differences between reference and acetic acids (computation may take some time)
  • The first 5 values of pa should be
> pa[1:5]
[1] 0.94898907 0.07179619 0.24797684 0.48204100 0.23177496
  • Run the test for all four acids
  • Report how many genes were significant with p-value <= 0.01 for each acid
  • Report how many genes are significant for both acetic and benzoate acids? (logical and is written as &)

Task C: multiple testing correction

Run the following snippet of code, which works on the vector of p-values pa obtained for acetate in task B

# adjusts vectors of p-vales from tasks B for using Bonferroni correction
pa.adjusted = p.adjust(pa, method ="bonferroni")
# add this adjusted vector to frame a
a <-  cbind(a, pa.adjusted)
# create permutation ordered by pa.adjusted
oa = order(pa.adjusted)
# select from table five rows with the lowest pa.adjusted (using vector oa)
# and display columns containing reference, acetate and adjusted p-value
a[oa[1:5],c(1:6,16)]

You should get output like this:

            ref1     ref2     ref3  acetate1   acetate2  acetate3 pa.adjusted
SUL1    7.581312 7.394985 7.412040 2.1633230 2.05412373 1.9169226 0.004793318
YMR244W 2.985682 2.975530 3.054001 0.3364722 0.33647224 0.1823216 0.188582576
DIP5    6.943991 7.147795 7.296955 0.6931472 0.09531018 0.5306283 0.253995075
YLR460C 5.620401 5.801212 5.502482 3.2425924 3.48431229 3.3843903 0.307639012
HXT4    2.821379 3.049273 2.772589 7.7893717 8.24446541 8.3041980 0.573813502

Do the same procedure for benzoate p-values and report the result. Comment the results for both acids.

Task D: volcano plot, test on data generated from null hypothesis

Draw a volcano plot for the acetate data

  • x-axis of this plot is the difference in the mean of reference and mean of acetate.
    • You can compute row means of a matrix by rowMeans.
  • y-axis is -log10 of the p-value (use original p-values before multiple testing correction)
  • You can quickly see genes which have low p-values (high on y-axis) and also big difference in mean expression between the two conditions (far from 0 on x-axis). You can also see if acetate increases or decreases expression of these genes.

Now create a simulated dataset sharing some features of the real data but observing the null hypothesis that the mean of reference and acetate are the same for each gene

  • Compute vector m of means for columns 1:6 from matrix a
  • Compute vectors sr and sa of standard deviations for reference columns and for acetate columns respectively
    • You can compute standard deviation for each row of a matrix by apply(some.matrix, 1, sd)
  • For each i in 1:6398, create three samples from normal distribution with mean m[i] and standard deviation sr[i], and three samples with mean m[i] and deviation sa[i]
    • Use function rnorm
  • On the resulting matrix apply Welch's t-test and draw the volcano plot.
  • How many random genes have p-value <=0.01? Is it roughly what we would expect under the null hypothesis?

Draw histogram of p-values from the real data (reference vs acetate) and from random data. Describe how they differ. Is it what you would expect?

  • use function hist

Submit plots volcano-real.png, volcano-random.png, hist-real.png, hist-random.png (real for real expression data and random for generated data)

L11

#HW11

Biological story: tiny monkeys

  • Common marmoset (Callithrix jacchus, Kosmáč bielofúzý) weights only about 1/4 kg
  • Most primates are much bigger
  • Which marmoset genes differ from other primates and are related to the small size?
  • Positive selection scan computes of each gene a p-value, whether it evolved on the marmoset lineage faster
    • Exact details, see papers [26] and [27]
  • The result is a list of p-values, one for each gene
  • Which biological functions are enriched among positively selected genes? Are any of those functions possibly related to body size?

Gene functions and GO categories

Use mysql database "marmoset" on the server.

  • We can look at the description of a particular gene:
select * from genes where prot='IGF1R';
+----------------------------+-------+-------------------------------------------------+                       
| transcriptid               | prot  | description                                     |                       
+----------------------------+-------+-------------------------------------------------+                       
| knownGene.uc010urq.1.1.inc | IGF1R | insulin-like growth factor 1 receptor precursor |                       
+----------------------------+-------+-------------------------------------------------+
  • In the database, we have stored all the P-values from positive selection tests:
select * from lrtmarmoset where transcriptid='knownGene.uc010urq.1.1.inc';
+----------------------------+---------------------+
| transcriptid               | pval                |
+----------------------------+---------------------+
| knownGene.uc010urq.1.1.inc | 0.00142731425252827 |
+----------------------------+---------------------+
  • Genes are also assigned functional categories based on automated processes (including sequence similarity to other genes) and manual curation. The corresponding database is maintained by Gene Ontology Consortium. We can use on-line sources to search for these annotations, e.g. here.
  • We can also download the whole database and preprocess it into usable form:
select * from genes2gocat,gocatdefs where transcriptid='knownGene.uc010urq.1.1.inc' and genes2gocat.cat=gocatdefs.cat;
(results in 50 categories)
  • GO categories have a hierarchical structure - see for example category GO:0005524 ATP binding:
select * from gocatparents,gocatdefs where gocatparents.parent=gocatdefs.cat and gocatparents.cat='GO:0005524';
+------------+------------+---------+------------+-------------------------------+
| cat        | parent     | reltype | cat        | def                           |
+------------+------------+---------+------------+-------------------------------+
| GO:0005524 | GO:0032559 | isa     | GO:0032559 | adenyl ribonucleotide binding |
+------------+------------+---------+------------+-------------------------------+
... and continuing further up the hierarchy:
| GO:0032559 | GO:0030554 | isa     | GO:0030554 | adenyl nucleotide binding     |
| GO:0032559 | GO:0032555 | isa     | GO:0032555 | purine ribonucleotide binding |
| GO:0030554 | GO:0001883 | isa     | GO:0001883 | purine nucleoside binding |
| GO:0030554 | GO:0017076 | isa     | GO:0017076 | purine nucleotide binding |
| GO:0032555 | GO:0017076 | isa     | GO:0017076 | purine nucleotide binding |
| GO:0032555 | GO:0032553 | isa     | GO:0032553 | ribonucleotide binding    |
| GO:0001883 | GO:0001882 | isa     | GO:0001882 | nucleoside binding |
| GO:0017076 | GO:0000166 | isa     | GO:0000166 | nucleotide binding |
| GO:0032553 | GO:0000166 | isa     | GO:0000166 | nucleotide binding |
| GO:0001882 | GO:0005488 | isa     | GO:0005488 | binding |
| GO:0000166 | GO:0005488 | isa     | GO:0005488 | binding |
| GO:0005488 | GO:0003674 | isa     | GO:0003674 | molecular_function |
  • What else can be under GO:0032559 adenyl ribonucleotide binding?
select * from gocatparents,gocatdefs where gocatparents.cat=gocatdefs.cat and gocatparents.parent='GO:0032559';
+------------+------------+---------+------------+-------------+
| cat        | parent     | reltype | cat        | def         |
+------------+------------+---------+------------+-------------+
| GO:0005524 | GO:0032559 | isa     | GO:0005524 | ATP binding |
| GO:0016208 | GO:0032559 | isa     | GO:0016208 | AMP binding |
| GO:0043531 | GO:0032559 | isa     | GO:0043531 | ADP binding |
+------------+------------+---------+------------+-------------+

Mann–Whitney U test

  • also known as Wilcoxon rank-sum test
  • In Lecture 10, we have used Welch's t-test to test if one set of expression measurements for a gene are significantly different from the second set
  • This test assumes that both sets come from normal (Gaussian) distributions with unknown parameters
  • Mann-Whitney U test is called non-parametric, because it does not make this assumption
  • The null hypothesis is that two sets of measurements were generated by the same unknown probability distribution
  • Alternative hypothesis: for X from the first distribution and Y from the second P(X>Y) is not equal P(Y>X)
  • We will use a one-side version of the alternative hypothesis: P(X>Y) > P(Y>X)
  • Compute test statistics U:
    • compare all pairs X, Y (X from first set, Y from second set)
    • if X>Y, add 1 to U
    • if X==Y, add 0.5
  • For large sets, U is approximately normally distributed under the null hypothesis

How to use in R:

# generate 20 samples from exponential distrib. with mean 1
x = rexp(20, 1)  
# generate 30 samples from exponential distrib. with mean 1/2
y = rexp(30, 2)  

# test if values of x greater than y
wilcox.test(x,y,alternative="greater")  
# W = 441, p-value = 0.002336
# alternative hypothesis: true location shift is greater than 0
# W is the U statistics above

# now generate y twice from the same distrib. as x
y = rexp(30, 1)
wilcox.test(x,y,alternative="greater")
# W = 364, p-value = 0.1053
# relatively small p-value (by chance)

y = rexp(30, 1)
wilcox.test(x,y,alternative="greater")
# W = 301, p-value = 0.4961
# now much greater p-value

Another form of the test, potentially useful for HW:

  • have a vector of values x, binary vector b indicating two classes: 0 and 1
  • test if values marked by 0 are greater than values marked by 1
# generate 10 with mean 1, 30 with mean 1/2, 10 with mean 1
x = c(rexp(10,1),rexp(30,2),rexp(10,1))
# classes 10x0, 20x1, 10x0
b = c(rep(0,10),rep(1,30),rep(0,10))
wilcox.test(x~b,alternative="greater")

# the same test by distributing into subvectors x0 and x1 for classes 0 and 1
x0 = x[b==0]
x1 = x[b==1]
wilcox.test(x0,x1,alternative="greater")
# should be the same as above

HW11

Lecture 11

  • In this task, you can use a combination of any scripting languages (e.g. Perl, Python, R) but also SQL, command-line tools etc.
  • Input is in a database
  • Submit required text files (optionally also files with figures in bonus part E)
  • Also submit any scripts you have written for this HW
  • In the protocol, include shell commands you have run
  • Outline of protocol is in /tasks/hw11/protocol.txt

Available data

  • All data necessary for this task is available in the mysql database 'marmoset' on the server
  • You will find password in /tasks/hw11/readme.txt
  • You have read-only access to the 'marmoset' database
  • For creating temporary tables, etc., you can use database 'temp_youruserid' (e.g. 'temp_mrkvicka54'), where you are allowed to create new tables and store data
  • You can address tables in mysql even between databases: to start client with your writeable database as default location, use:
mysql -p temp_mrkvicka54
  • You can then access data in the table 'genes' in the database 'marmoset' simply by using 'marmoset.genes'

Getting data from database:

  • If your want to get data from database to a tab-separated file, write a select query, run with -e, redirect output:
mysql -p marmoset -e 'select transcriptid as id, pval from lrtmarmoset' > pvals.tsv

Task A: Find ancestors of each GO category

  • Compute a table (in your temporary db or in a file) which contains all pairs category and its ancestor
    • In table gocatparents you have pairs category, its parent, so you need a transitive closure over this relation
    • SQL is not very good at this, you can try repeated joins until you find no more ancestors
    • Alternatively, you can simply extract data from the database and process them in a language of your choice
  • Submit file sample-anc.txt which contains the list of all ancestors of GO:0042773, one per line, in sorted order
    • There should be 14 such ancestors, excluding this category itself; the first in sorted order is GO:0006091, the last is GO:0055114

Task B: Gene in/out data for each GO category

  • Again consider category GO:0042773
  • Create a list of all genes that occur in table lrtmarmoset
    • for each such gene list three columns separated by tabs: its transcript id, p-value from lrtmarmoset, and an indicator 0/1
    • the indicator is 1, if this gene occurs in GO:0042773 or one of its subcategories; 0 otherwise
    • to find, which gene occur directly in GO:0042773, use table genes2gocat, subcategories can be found in your table from part A
    • note that genes2gocat contains more genes, we will consider only genes from lrtmarmoset
  • Submit this file sample-genes.tsv
    • Overall, your table should have 13717 genes, out of which 28 have value 1 in the last column
    • The first lines of this list (when sorted alphabetically) might look as follos:
ensGene.ENST00000043410.1.inc   1       0
ensGene.ENST00000158526.1.inc   0.483315913388483       0
...
ensGene.ENST00000456284.1       1       1
...
  • Note that in part C, you will need to run this process for each category in the database, so make it sufficiently automated

Task C: Run Man-Whitney U test for each GO category

  • Run Man-Whitney U test for each non-trivial category
    • Non-trivial categories are such that at least one of our genes (from lrtmarmoset) is in the category and at least one of our genes is not in the category
    • You should test, if genes in a particular GO category have smaller p-values in positive selection than genes outside the category
    • List of all categories can be obtained from gocatdefs, but not all of them are non-trivial (there are 12455 non-trivial categories)
  • Submit file test.tsv in which each line contains two tab separated values:
    • GO category id
    • p-value from the test
  • For partial points test at least the category GO:0042773 from parts A and B

Task D: Report significant categories

  • Submit file report.tsv with 20 most significant GO categories (lowest p-values)
    • For each category list its ID, p-value and description
    • Order them from the most significant
    • Descriptions are in table gocatdefs
  • To your protocol, write any observations you can make
    • Do any reported categories seem interesting to you?
    • Are any reported categories likely related to each other based on their descriptions?

Task E (bonus): cluster significant categories

  • Some categories in task D appear similar according to their name
  • Try creating k-means or hierarchical clustering of categories
  • Represent each category as a binary vector in which for each gene you have one bit indicating if it is in the category
  • Thus categories with the same set of genes will have identical vectors
  • Try to report results in an appropriate form (table, text, figure), discuss them in the protocol

Note

  • In part C, we have done many statistical tests, resulting P-values should be corrected by multiple testing correction from Lecture 10
    • This is not required in this homework, but should be done in a real study