Mercurial > repos > antmarge > compregions
view compRegions.pl @ 2:510317f64293 draft
Deleted selected files
author | antmarge |
---|---|
date | Tue, 28 Mar 2017 21:51:48 -0400 |
parents | 198cfbcae096 |
children |
line wrap: on
line source
#!/usr/bin/perl -w #Margaret Antonio 17.01.10 For GALAXY use Data::Dumper; use strict; use Getopt::Long; use Statistics::Distributions; no warnings; #ASSIGN INPUTS TO VARIABLES USING FLAGS our ($input1, $input2, $help,$out,$round,$l1, $l2); GetOptions( 'input1:s' => \$input1, 'input2:s' => \$input2, 'h' => \$help, 'o:s' =>\$out, 'r:i'=> \$round, 'l1:s'=> \$l1, 'l2:s'=> \$l2, ); sub print_usage() { print "\n####################################################################\n"; print "\n"; print "compWindows.pl: compare regions outputted by the slidingWindow tool\n\n"; print "DESCRIPTION: Takes two slidingWindows.csv files and compares them by\n"; print "calculating the difference in mean fitness, the pval for each gene.\n"; print "Example: compare control vs antibiotic, where both from same strain (genome).\n"; print "Note: For different strains/genomes, use compStrains for a gene comparison.pl\n"; print "\nUSAGE:\n"; print "perl compGenes.pl -d inputs/ -l cond1,cond2 -r 2 -o comp-cond1cond2_date.csv\n"; print "\nREQUIRED:\n"; print " -d\tDirectory containing two slidingWindow.csv files (output of\n"; print " \tslidingWindow tool)\n"; print " \tOR\n"; print " \tIn the command line (without a flag), input the name(s) of\n"; print " \ttwo files.\n"; print "\nOPTIONAL:\n"; print " -h\tPrints usage and exits program\n"; print " -o\tOutput file for comparison data. Default: label1label2.csv\n"; print " -r\tRound final output numbers to this number of decimals\n"; print " -s\tColumn number to sort by. Default: difference of means\n"; print " -l\tLabels for input files. Default: filenames\n"; print " \tTwo strings, comma separated (i.e. -l expt1,expt2).\n"; print " \tOrder should match file order.\n"; print " \n~~~~Always check that file paths are correctly specified~~~~\n"; print "\n##################################################################\n"; } if ($help){ print_usage(); exit; } #THE @files ARRAY WILL CONTAIN INPUT FILE NAMES my @files=($input1,$input2); #GET LABELS: USE (-l) OR USE FILNEAMES AS LABELS FOR COLUMNS IN OUTPUT FILE my @labels=($l1,$l2); #INDICES 0:start 1:end 2:mutants 3:insertions 4:TA_sites 5:ratio 6:p-value #7:average 8:variance 9:stdev 10:stderr 11:genes 12:fit-mean #CHECK IF REQ. VARIABLES WERE DEFINED USING FLAGS. IF NOT THEN USE DEFAULT VALUES if (!$out) {$out="comp.".$labels[0].$labels[1].".csv"} if (!$round){$round='%.4f'} #READ INPUT FILES AND PUT GENE INFO FROM EACH LINE INTO HASH %all WHERE KEY=GENE_ID AND #VALUE=GENE INFO (MEAN FITNESS, # INSERTIONS, ETC.) my %all; my @header; for (my $i=0; $i<2; $i++){ print "File #",$i+1,"\t"; my $file=$files[$i]; print $file,"\n"; open(DATA, '<', $file) or (print "Could not open '$file'\n" and print_usage() and exit); #EXTRACT THE HEADER (COLUMN NAMES) OF THE FILE AND KEEP FOR OUTPUT HEADER #APPEND FILE NAME OR GIVEN LABEL (-l) TO COLUMN NAME SO ITS DIFFERENT FROM OTHER INPUT FILE my $head=<DATA>; my @cols=split(',',$head); for (my $j=0;$j<scalar @cols;$j++){ my $colname=$cols[$j]; chomp($colname); $cols[$j]=$colname.'-'.$labels[$i]; print $cols[$j],"\n"; } push (@header,@cols); while (my $entry = <DATA>) { chomp $entry; my @line=split(",",$entry); if (!$line[11]){ $line[11]="NA"; } my ($start,$end,$mutants,$insertions,$TAsites,$ratio,$pval,$avg,$var,$sd,$se,$genes)=@line; $avg=sprintf("%.3f",$avg); #if (!$line[6]){ # $line[6]=0; #} #@line=@line[0,1,2,3,4,5,6]; #PUT GENE AND INFO INTO THE HASH FOR EXISTING KEY (1ST FILE) OR CREATE NEW KEY (2ND FILE) my $key=$start; if(!exists $all{$key}){ my @info; push (@info,@line); $all{$key}=\@info; } #Otherwise the window existed for the prior file, and now need to calcualte extra stats else{ my @info=@{$all{$key}}; my ($start2,$end2,$mutants2,$insertions2,$TAsites2,$ratio2,$pval2,$avg2,$var2,$sd2,$se2,$genes2)=@info; my $diff=sprintf("$round",($avg-$avg2)); my $df=$insertions+$insertions2-2; my $tdist; my $pval; # CHECK TO MAKE SURE ALL VARIABLES IN TDIST,PVAL CALCULATIONS ARE NUMBERS AND NO # ZEROS (0) IN THE DENOMINATOR if ($se eq "NA" or $se2 eq "NA" or $sd eq "NA" or $sd2 eq "NA" or $insertions==0 or $insertions2==0 or $sd==0 or $sd2==0 or $df<=0){ ($tdist,$pval)=(" "," "); } else{ $tdist=sqrt((($diff)/(sqrt((($sd**2)/$insertions)+(($sd2**2)/$insertions2))))**2); $pval=Statistics::Distributions::tprob($df,$tdist); } push (@info,@line,$diff,$df,$tdist,$pval); $all{$key}=\@info; } } close DATA; } #READ ALL HASH CONTENTS INTO 2D ARRAY FOR EASY SORTING AND PRINTING TO OUT FILE my @unsorted; foreach my $entry (keys %all) { my @info=@{$all{$entry}}; my @temp; push (@temp,@info); push (@unsorted,\@temp); } #SORT GENES BY PVALUE OR FITNESS DEPENDING ON WHICH FLAG WAS USED IN COMMANDLINE my $sortkey=26; #default: sort by difference of means my @sorted = sort { $b->[$sortkey] <=> $a->[$sortkey] } @unsorted; #ADD NEW FIELDS TO HEADER (COLUMN NAMES) my $field="Mean".$labels[0].'.'.$labels[1]; push (@header,$field,"DOF","TDIST","PVALUE"); #PRINT TO OUT FILE open OUT, '>',"$out"; print OUT (join(',',@header),"\n"); foreach (@sorted){ my @woo=@{$_}; print OUT join(',',@woo),"\n"; } close OUT;