a21f21b13b09a794b09e89168a51d4c8a1da85af max Wed Sep 14 11:47:52 2016 -0700 finally committing all missing pieces of the crispr track pipeline, refs #17235 diff --git src/utils/doCrispr.sh src/utils/doCrispr.sh index 35f7c47..c3570fd 100755 --- src/utils/doCrispr.sh +++ src/utils/doCrispr.sh @@ -1,131 +1,133 @@ #/bin/bash set -euo pipefail # script to create a crispr track for an assembly # has only two arguments: the genome db and the gene track to use # steps: # - get 200bp-regions around exons # - get the 20bp-GG sequences in them # - take only the unique ones and split these into chunks set -e set -o pipefail # path to a git clone of the crispor github repo, must be under /hive -CRISPOR=/hive/users/max/projects/crispor/crispor +CRISPOR=/hive/data/outside/crisprTrack/crispor if [ ! -f $CRISPOR/crispor.py ]; then echo error: cannot find $CRISPOR/crispor.py exit 1 fi -# path to the crispr track tools, should be on /hive, from kent/src/utils/crisprTrack -crisprTrack=/hive/groups/browser/crisprTrack +# path to the crispr track pipeline scripts +# look at ~/kent/src/hg/makeDb/crisprTrack/README.txt for +# more info about these scripts in there +crisprTrack= /hive/data/outside/crisprTrack/scripts db=$1 geneTrackName=$2 chromSizesPath=/hive/data/genomes/$db/chrom.sizes if [[ -z "$1" || -z "$2" ]]; then echo to build a crispr track, you need to specify two arguments: echo 1 - genome db echo 2 - name of a gene track exit 1 fi genomeFname=$CRISPOR/genomes/$db/$db.fa.bwt if [ ! -f $genomeFname ]; then echo error: $genomeFname not found - make sure that you have bwa indexed this genome in crispor exit 1 fi mkdir /hive/data/genomes/$db/bed/crispr cd /hive/data/genomes/$db/bed/crispr echo `date` $USER >> doCrispr.log echo gene track: $geneTrackName >> doCrispr.log mkdir -p ranges # select fields explicitly: some dbs have bin fields, others don't echo getting genes hgsql $db -NB -e 'select name,chrom,strand,txStart,txEnd,cdsStart,cdsEnd,exonCount,exonStarts,exonEnds from '$geneTrackName' where chrom not like "%_alt" and chrom not like "%hap%"; ' > ranges/genes.gp echo Number of transcripts: >> doCrispr.log echo `wc -l ranges/genes.gp` >> doCrispr.log echo break genes into exons and add 200 bp on each side genePredToBed ranges/genes.gp stdout | grep -v hap | grep -v chrUn | bedToExons stdin stdout | awk '{$2=$2-200; $3=$3+200; $6="+"; print}'| bedClip stdin /hive/data/genomes/$db/chrom.sizes stdout | grep -v _alt | grep -i -v hap > ranges/ranges.bed echo get sequence. this can take 10-15 minutes due to our old twoBit.c bug echo featureBits of target ranges >> doCrispr.log featureBits $db ranges/ranges.bed -faMerge -fa=ranges/ranges.fa -minSize=20 -bed=stdout | cut -f-3 2>> doCrispr.log > crisprRanges.bed echo split the sequence file into pieces for the cluster mkdir -p ranges/inFa/ ranges/outGuides # split sequences into smaller parts faSplit sequence ranges/ranges.fa 100 ranges/inFa/ echo find all guides in the pieces # now pull out all potential guide sequences within these ranges in .bed and .fa # don't know how to do $var expansion in heredoc, so using simple loop, not template/gensub2 for i in ranges/inFa/*.fa; do echo python $crisprTrack/findGuides.py {check in exists inFa/`basename $i`} {check in exists $chromSizesPath} outGuides/`basename $i .fa`.bed {check out exists outGuides/`basename $i`}; done > ranges/jobList # actually just a few minutes on the cluster ssh ku "cd `pwd`/ranges; para freeBatch; para make jobList" echo concat all guides that were found cat ranges/outGuides/*.fa | grep -v \> > allGuides.txt cat ranges/outGuides/*.bed > allGuides.bed echo preparing the biggest cluster job, specificity alignment: echo filtering and splitting guide sequences into small pieces mkdir -p specScores/jobs/inFa specScores/jobs/outGuides specScores/jobs/outOffs # the next step will take 30 mins or so # It will create 1 million tiny files for the cluster jobs. # splitGuidesSpecScore.py will remove all non-unique sequences from allGuides.txt # it will also re-format the guide sequences into the weird format required by the CRISPOR # script: separate them by NNNN and append the artificial AGG PAM, so CRISPOR accepts # these sequences and finds off-targets for them. The NNNN are needed so CRISPOR won't find # any guides in overlaps of two guide sequences, as CRISPOR ignores all guides with Ns. # however, CRISPOR will still find guides in all sequences that start with CC, which # will waste a bit of CPU. python $crisprTrack/splitGuidesSpecScore.py allGuides.txt specScores/jobs/inFa specScores/jobNames.txt echo creating a jobList file from the pieces # make the big jobList file, I used python, but not really necessary python $crisprTrack/makeSpecJoblist.py specScores/jobNames.txt $CRISPOR $db specScores/jobList ssh ku "cd `pwd`/specScores; para freeBatch; para make jobList" # also calc the efficiency scores python $crisprTrack/splitGuidesEffScore.py $chromSizesPath allGuides.bed effScores/jobs effScores/jobNames.txt # cannot get $db substitution to work with a bash heredoc for i in `find effScores/jobs/bed -type f`; do echo /cluster/software/bin/python $crisprTrack/jobCalcEffScores.py $CRISPOR $db {check in exists jobs/bed/`basename $i`} {check out exists jobs/out/`basename $i .bed`.tab}; done > effScores/jobList ssh ku "cd `pwd`/effScores; para freeBatch; para make jobList" # now concat the cluster job output back into two files # around 10 mins each find specScores/jobs/outGuides -type f | xargs cut -f-4 > specScores.tab find effScores/jobs/out/ -type f | xargs cat > effScores.tab echo Number of guides >> doCrispr.log echo `wc -l specScores.tab` >> doCrispr.log echo Number of guides >> doCrispr.log echo `wc -l specScores.tab` >> doCrispr.log echo converting off-targets mkdir specScores/catJobs/inFnames specScores/catJobs/out/ -p echo specScores/jobs/outOffs/*.tab | tr ' ' '\n' | sed -e s/specScores/../g > specScores/otFnames.txt splitFile specScores/otFnames.txt 20 specScores/catJobs/inFnames/otJob for i in specScores/catJobs/inFnames/otJob*; do fname=`basename $i`; echo python /hive/groups/browser/crisprTrack/convOffs.py $db {check in exists inFnames/$fname} {check out exists out/$fname}; done > specScores/catJobs/jobList echo concating and indexing the off-targets catAndIndex specScores/catJobs/out offtargets.tab offtargets.offsets.tab # create the bigBed file # approx 30 mins on human echo creating a bigBed file time python $crisprTrack/createBigBed.py $db allGuides.bed specScores.tab effScores.tab offtargets.offsets.tab ln -s `pwd`/crispr.bb /gbdb/$db/bbi/crispr.bb hgBbiDbLink $db crisprTargets /gbdb/$db/bbi/crispr.bb hgLoadBed $db crisprRanges crisprRanges.bed