Mercurial > repos > arkarachai-fungtammasan > str_fm
view commandline_sample_STR-FM_shortread_profiling @ 4:d74dc58cbd53 draft
Uploaded
author | arkarachai-fungtammasan |
---|---|
date | Sat, 22 Aug 2015 12:12:55 -0400 |
parents | d5ed5c2e25c3 |
children |
line wrap: on
line source
## This is a sample PBS script for profiling STR from short read using STR-FM version 2.0.0 (April 20, 2015) ## ##requirement ##1 fastq input in sangerfq Phred scale --> ${INPUT}.fastq ##2 index of mapping program (bwa, bowtie, etc) ##3 location of all STR in reference genome (use PBS script name "sampleSTR_reference_profiling.txt) --> /path/to/STR/in/reference/genome.TR (you can make 4 separated TR files for 4 types of STRs) ##4 reference genome in FASTA and in 2bit file --> /path/to/2bit/ref.2bit (use utility from UCSC genome browser to create 2bit file version of reference genome) ##5 local Galaxy (available from Galaxy website for Mac and Unix computer) ##6 STR error rates (can be downloaded from https://usegalaxy.org/u/guru%40psu.edu/h/error-rates-files) --> errorrate.bymajorallele ## echo " " echo " " echo "Job started on `hostname` at `date`" ref=/path/to/reference/sequence/and/bwa/index/ref.fa export PYTHONPATH=/path/to/galaxy-dist/lib/ galaxydir=/path/to/galaxy-dist/tools cd /working/directory/ echo " " echo " detect STR in short read" ## See detail in microsatellite.xml on https://github.com/Arkarachai/STR-FM python microsatellite.py ${INPUT}.fastq --fastq --period=1 --partialmotifs --minlength=5 --prefix=20 --suffix=20 --hamming=0 --multipleruns >${INPUT}.mono.out python microsatellite.py ${INPUT}.fastq --fastq --period=2 --partialmotifs --minlength=6 --prefix=20 --suffix=20 --hamming=0 --multipleruns >${INPUT}.di.out python microsatellite.py ${INPUT}.fastq --fastq --period=3 --partialmotifs --minlength=9 --prefix=20 --suffix=20 --hamming=0 --multipleruns >${INPUT}.tri.out python microsatellite.py ${INPUT}.fastq --fastq --period=4 --partialmotifs --minlength=12 --prefix=20 --suffix=20 --hamming=0 --multipleruns >${INPUT}.tetra.out echo "change read name at " ## See detail in space2underscore_readname.xml on https://github.com/Arkarachai/STR-FM python changespacetounderscore_readname.py ${INPUT}.mono.out ${INPUT}.mono.new 6 python changespacetounderscore_readname.py ${INPUT}.di.out ${INPUT}.di.new 6 python changespacetounderscore_readname.py ${INPUT}.tri.out ${INPUT}.tri.new 6 python changespacetounderscore_readname.py ${INPUT}.tetra.out ${INPUT}.tetra.new 6 echo "start fetch flanking at `date`" ## See detail in fetchflank.xml on https://github.com/Arkarachai/STR-FM python pair_fetch_DNA_ff.py ${INPUT}.mono.new ${INPUT}.mono_ff_L.txt ${INPUT}.mono_ff_R.txt 20 20 python pair_fetch_DNA_ff.py ${INPUT}.di.new ${INPUT}.di_ff_L.txt ${INPUT}.di_ff_R.txt 20 20 python pair_fetch_DNA_ff.py ${INPUT}.tri.new ${INPUT}.tri_ff_L.txt ${INPUT}.tri_ff_R.txt 20 20 python pair_fetch_DNA_ff.py ${INPUT}.tetra.new ${INPUT}.tetra_ff_L.txt ${INPUT}.tetra_ff_R.txt 20 20 echo "BWA uniquely mapped no indel no deletion " bwa aln -n 0 -o 0 ${ref} ${INPUT}.mono_ff_L.txt > ${INPUT}.mono_ff_L.sai bwa aln -n 0 -o 0 ${ref} ${INPUT}.mono_ff_R.txt > ${INPUT}.mono_ff_R.sai bwa sampe ${ref} ${INPUT}.mono_ff_L.sai ${INPUT}.mono_ff_R.sai ${INPUT}.mono_ff_L.txt ${INPUT}.mono_ff_R.txt > ${INPUT}.mono.sam samtools view -Sb -F 12 -q 1 ${INPUT}.mono.sam > ${INPUT}.mono.n.all.bam bwa aln -n 0 -o 0 ${ref} ${INPUT}.di_ff_L.txt > ${INPUT}.di_ff_L.sai bwa aln -n 0 -o 0 ${ref} ${INPUT}.di_ff_R.txt > ${INPUT}.di_ff_R.sai bwa sampe ${ref} ${INPUT}.di_ff_L.sai ${INPUT}.di_ff_R.sai ${INPUT}.di_ff_L.txt ${INPUT}.di_ff_R.txt > ${INPUT}.di.sam samtools view -Sb -F 12 -q 1 ${INPUT}.di.sam > ${INPUT}.di.n.all.bam bwa aln -n 0 -o 0 ${ref} ${INPUT}.tri_ff_L.txt > ${INPUT}.tri_ff_L.sai bwa aln -n 0 -o 0 ${ref} ${INPUT}.tri_ff_R.txt > ${INPUT}.tri_ff_R.sai bwa sampe ${ref} ${INPUT}.tri_ff_L.sai ${INPUT}.tri_ff_R.sai ${INPUT}.tri_ff_L.txt ${INPUT}.tri_ff_R.txt > ${INPUT}.tri.sam samtools view -Sb -F 12 -q 1 ${INPUT}.tri.sam > ${INPUT}.tri.n.all.bam bwa aln -n 0 -o 0 ${ref} ${INPUT}.tetra_ff_L.txt > ${INPUT}.tetra_ff_L.sai bwa aln -n 0 -o 0 ${ref} ${INPUT}.tetra_ff_R.txt > ${INPUT}.tetra_ff_R.sai bwa sampe ${ref} ${INPUT}.tetra_ff_L.sai ${INPUT}.tetra_ff_R.sai ${INPUT}.tetra_ff_L.txt ${INPUT}.tetra_ff_R.txt > ${INPUT}.tetra.sam samtools view -Sb -F 12 -q 1 ${INPUT}.tetra.sam > ${INPUT}.tetra.n.all.bam echo "sort result by read name" samtools sort -n ${INPUT}.mono.n.all.bam ${INPUT}.mono.n.sorted.all samtools sort -n ${INPUT}.di.n.all.bam ${INPUT}.di.n.sorted.all samtools sort -n ${INPUT}.tri.n.all.bam ${INPUT}.tri.n.sorted.all samtools sort -n ${INPUT}.tetra.n.all.bam ${INPUT}.tetra.n.sorted.all samtools view -h -o ${INPUT}.mono.n.sorted.all.sam ${INPUT}.mono.n.sorted.all.bam samtools view -h -o ${INPUT}.di.n.sorted.all.sam ${INPUT}.di.n.sorted.all.bam samtools view -h -o ${INPUT}.tri.n.sorted.all.sam ${INPUT}.tri.n.sorted.all.bam samtools view -h -o ${INPUT}.tetra.n.sorted.all.sam ${INPUT}.tetra.n.sorted.all.bam echo "merge faux paired end reads" ## See detail in PEsortedSAM2readprofile.xml on https://github.com/Arkarachai/STR-FM python PEsortedSAM2readprofile.py ${INPUT}.mono.n.sorted.all.sam /path/to/2bit/ref.2bit 100 250 ${INPUT}.mono.RF python PEsortedSAM2readprofile.py ${INPUT}.di.n.sorted.all.sam /path/to/2bit/ref.2bit 100 250 ${INPUT}.mono.RF python PEsortedSAM2readprofile.py ${INPUT}.tri.n.sorted.all.sam /path/to/2bit/ref.2bit 100 250 ${INPUT}.mono.RF python PEsortedSAM2readprofile.py ${INPUT}.tetra.n.sorted.all.sam /path/to/2bit/ref.2bit 100 250 ${INPUT}.mono.RF echo "join mapped coordinate with STR length using read name" python ${galaxydir}/filters/join.py ${INPUT}.mono.new ${INPUT}.mono.RF 6 1 ${INPUT}.mono.RF.j "" "" --index_depth=3 --buffer=50000000 --fill_options_file='None' python ${galaxydir}/filters/join.py ${INPUT}.di.new ${INPUT}.di.RF 6 1 ${INPUT}.mono.RF.j "" "" --index_depth=3 --buffer=50000000 --fill_options_file='None' python ${galaxydir}/filters/join.py ${INPUT}.tri.new ${INPUT}.tri.RF 6 1 ${INPUT}.mono.RF.j "" "" --index_depth=3 --buffer=50000000 --fill_options_file='None' python ${galaxydir}/filters/join.py ${INPUT}.tetra.new ${INPUT}.tetra.RF 6 1 ${INPUT}.mono.RF.j "" "" --index_depth=3 --buffer=50000000 --fill_options_file='None' echo "join mapped coordinate and STR length with STR location in genome" python ${galaxydir}/new_operations/gops_join.py /path/to/STR/in/reference/genome.TR ${INPUT}.mono.RF.j ${INPUT}.mono.gop -1 1,2,3,0 -2 10,13,14,0 -m 1 -f python ${galaxydir}/new_operations/gops_join.py /path/to/STR/in/reference/genome.TR ${INPUT}.di.RF.j ${INPUT}.di.gop -1 1,2,3,0 -2 10,13,14,0 -m 1 -f python ${galaxydir}/new_operations/gops_join.py /path/to/STR/in/reference/genome.TR ${INPUT}.tri.RF.j ${INPUT}.tri.gop -1 1,2,3,0 -2 10,13,14,0 -m 1 -f python ${galaxydir}/new_operations/gops_join.py /path/to/STR/in/reference/genome.TR ${INPUT}.tetra.RF.j ${INPUT}.tetra.gop -1 1,2,3,0 -2 10,13,14,0 -m 1 -f echo "remove incompatible motif (remove incorrect mapped reads given that there is no STR motif difference from reference genome)" ## See detail in microsatcompat.xml on https://github.com/Arkarachai/STR-FM python microsatcompat.py ${INPUT}.mono.gop 4 10 > ${INPUT}.mono.fulltable1 python microsatcompat.py ${INPUT}.di.gop 4 10 > ${INPUT}.di.fulltable1 python microsatcompat.py ${INPUT}.tri.gop 4 10 > ${INPUT}.tri.fulltable1 python microsatcompat.py ${INPUT}.tetra.gop 4 10 > ${INPUT}.tetra.fulltable1 echo "remove shifting flanking location (remove cases that come from STR interruption or flanking bases are misread as STRs)" cat ${INPUT}.mono.fulltable1 | awk '($19==$2) && ($20==$3) {print $0}' > ${INPUT}.mono.fulltable2 cat ${INPUT}.di.fulltable1 | awk '($19==$2) && ($20==$3) {print $0}' > ${INPUT}.di.fulltable2 cat ${INPUT}.tri.fulltable1 | awk '($19==$2) && ($20==$3) {print $0}' > ${INPUT}.tri.fulltable2 cat ${INPUT}.tetra.fulltable1 | awk '($19==$2) && ($20==$3) {print $0}' > ${INPUT}.tetra.fulltable2 echo "keep only column that are necessary for profiling" cat ${INPUT}.mono.fulltable2| cut -f 1,2,3,4,5,7 | sort -k 1n,1 -k 2n,2 -k 3n,3 > ${INPUT}.mono.cuttmp0 cat ${INPUT}.di.fulltable2| cut -f 1,2,3,4,5,7 | sort -k 1n,1 -k 2n,2 -k 3n,3 > ${INPUT}.di.cuttmp0 cat ${INPUT}.tri.fulltable2| cut -f 1,2,3,4,5,7 | sort -k 1n,1 -k 2n,2 -k 3n,3 > ${INPUT}.tri.cuttmp0 cat ${INPUT}.tetra.fulltable2| cut -f 1,2,3,4,5,7 | sort -k 1n,1 -k 2n,2 -k 3n,3 > ${INPUT}.tetra.cuttmp0 echo "If you multiple analysis by splitting initial fastq, you should merge (cat) all results from the same sample after this step" echo "create genomic coordinate column and group by that column" perl ${galaxydir}/filters/fixedValueColumn.pl ${INPUT}.mono.cuttmp0 ${INPUT}.mono.cuttmp1 "_" "no" python ${galaxydir}/filters/mergeCols.py ${INPUT}.mono.cuttmp1 ${INPUT}.mono.cuttmp2 1 7 2 7 3 python ${galaxydir}/stats/grouping.py ${INPUT}.mono.cuttmp3 ${INPUT}.mono.cuttmp2 8 0 'cat 6 0' 'cat_uniq 4 0' perl ${galaxydir}/filters/fixedValueColumn.pl ${INPUT}.di.cuttmp0 ${INPUT}.di.cuttmp1 "_" "no" python ${galaxydir}/filters/mergeCols.py ${INPUT}.di.cuttmp1 ${INPUT}.di.cuttmp2 1 7 2 7 3 python ${galaxydir}/stats/grouping.py ${INPUT}.di.cuttmp3 ${INPUT}.di.cuttmp2 8 0 'cat 6 0' 'cat_uniq 4 0' perl ${galaxydir}/filters/fixedValueColumn.pl ${INPUT}.tri.cuttmp0 ${INPUT}.tri.cuttmp1 "_" "no" python ${galaxydir}/filters/mergeCols.py ${INPUT}.tri.cuttmp1 ${INPUT}.tri.cuttmp2 1 7 2 7 3 python ${galaxydir}/stats/grouping.py ${INPUT}.tri.cuttmp3 ${INPUT}.tri.cuttmp2 8 0 'cat 6 0' 'cat_uniq 4 0' perl ${galaxydir}/filters/fixedValueColumn.pl ${INPUT}.tetra.cuttmp0 ${INPUT}.tetra.cuttmp1 "_" "no" python ${galaxydir}/filters/mergeCols.py ${INPUT}.tetra.cuttmp1 ${INPUT}.tetra.cuttmp2 1 7 2 7 3 python ${galaxydir}/stats/grouping.py ${INPUT}.tetra.cuttmp3 ${INPUT}.tetra.cuttmp2 8 0 'cat 6 0' 'cat_uniq 4 0' echo "you may filter for minimum sequencing depth here" echo "genotyping using error correction model" ## See detail in GenotypingSTR.xml on https://github.com/Arkarachai/STR-FM cat ${INPUT}.mono.cuttmp2 ${INPUT}.di.cuttmp2 ${INPUT}.tri.cuttmp2 ${INPUT}.tetra.cuttmp2 > ${INPUT}.step5 python GenotypeTRcorrection.py ${INPUT}.step5 errorrate.bymajorallele ${INPUT}.step5.result 0.5 ## final output is ${INPUT}.step5.result echo "Job end on `hostname` at `date`"