comparison kappa_report @ 0:2d601bd04c93 default tip

initial commit

0:2d601bd04c93

#!/usr/bin/env perl

use strict;
use warnings;


my $quiet = 0;
if(@ARGV and $ARGV[0] =~ /^-q/){
  $quiet = 1;
  shift @ARGV;
} 
# Calculates Cohen's kappa measure for inter-annotator agreement, here for two genotype callers
# SPECIAL ATTENTION MUST BE PAID to include or exclude calls using depth of coverage information, and only look at targeted regions (e.g. exome analysis)
# Also, the probability of a non-call must be incorporated, as there is going to be 99.9% agreement by default due to the low rate of 
# SNPs in the genome in human, for example.
@ARGV == 9 or die "Usage: $0 [-q(uiet)] <exome targets.bed> <coding regions.gtf> <genotypes1.hgvs.txt> <bam1> <coverage cutoff1> <genotypes2.hgvs.txt> <bam2> <coverage cutoff2> <output report.txt>\n";

my $targets_file = shift @ARGV;
my $coding_file = shift @ARGV;
my $calls1_file = shift @ARGV;
my $bam_file1 = shift @ARGV;
my $calls1_min_depth = shift @ARGV;
my $calls2_file = shift @ARGV;
my $bam_file2 = shift @ARGV;
my $calls2_min_depth = shift @ARGV;
my $output_file = shift @ARGV;

my %reporting;
print STDERR "Reading in target regions list...\n" unless $quiet;
open(BED, $targets_file)
  or die "Cannot open target BED file $targets_file for reading: $!\n";
while(<BED>){
    next if /^(\s*#|track=)/;
    chomp;
    my @fields = split /\t/, $_;

    if(not exists $reporting{$fields[0]}){
      $reporting{$fields[0]} = [[$fields[1], $fields[2]]];
      next;
    }
    push @{$reporting{$fields[0]}}, [$fields[1], $fields[2]];
}
close(BED);
for my $c (keys %reporting){
  $reporting{$c} = [sort {$a->[0] <=> $b->[0]} @{$reporting{$c}}];
}

my %coding;
print STDERR "Reading in coding regions list...\n" unless $quiet;
open(GTF, $coding_file)
  or die "Cannot open $coding_file for reading: $!\n";
while(<GTF>){
    next if /^\s*#/;
    my @fields = split /\t/, $_;

    if($fields[2] eq "CDS"){
      if(not exists $coding{$fields[0]}){
        $coding{$fields[0]} = [[$fields[3], $fields[4]]];
        next;
      }
      push @{$coding{$fields[0]}}, [$fields[3], $fields[4]];
    }
}
close(GTF);
for my $c (keys %coding){
  $coding{$c} = [sort {$a->[0] <=> $b->[0]} @{$coding{$c}}];
}

print STDERR "Reading in genotype calls...\n" unless $quiet;
open(OUT, ">$output_file")
  or die "Cannot open $output_file for writing: $!\n";
my ($calls1, $subpar_calls1) = retrieve_calls($calls1_file, \%reporting, \%coding, $calls1_min_depth);
my ($calls2, $subpar_calls2) = retrieve_calls($calls2_file, \%reporting, \%coding, $calls2_min_depth);

my $total_calls1 = 0;
for my $chr (keys %$calls1){
  $total_calls1 += scalar(keys %{$calls1->{$chr}});
}
my $total_calls2 = 0;
for my $chr (keys %$calls2){
  $total_calls2 += scalar(keys %{$calls2->{$chr}});
}

print OUT "Total Method 1 calls\t$total_calls1\n";
print OUT "Total Method 2 calls\t$total_calls2\n";
print STDERR "Comparing genotype calls...\n" unless $quiet;
# Go through the BAM file in all targeted locations to see if there is a reference or variant call (HGVS file only shows variants)
my $bases_diff = 0;
my $bases_same = 0;
my $zygosities_diff = 0;
my $zygosities_same = 0;
for my $chr (keys %$calls1){
  for my $pos (keys %{$calls1->{$chr}}){
    if(exists $calls2->{$chr}->{$pos}){
      # same variant base
      if($calls2->{$chr}->{$pos}->[0] eq $calls1->{$chr}->{$pos}->[0]){
        $bases_same++;
        if($calls1->{$chr}->{$pos}->[1] eq $calls2->{$chr}->{$pos}->[1]){
          $zygosities_same++;
        }
        else{
          $zygosities_diff++;
        }
      }
      # diff variant base
      else{
        $bases_diff++;
      }
      delete $calls1->{$chr}->{$pos}; # so in the end all that's left is method-1-only calls in the hash 
      delete $calls2->{$chr}->{$pos}; # so in the end all that's left is method-2-only calls in the hash 
      delete $subpar_calls1->{$chr}->{$pos} if exists $subpar_calls1->{$chr}; # so in the end all that's left is method-1-only calls in the hash 
      delete $subpar_calls2->{$chr}->{$pos} if exists $subpar_calls2->{$chr}; # so in the end all that's left is method-2-only calls in the hash 
    }
  }
}

print OUT "Same calls\t$zygosities_same\n";
print OUT "Diff zygosity\t$zygosities_diff\n";
print OUT "Diff bases\t$bases_diff\n";

my $covbed = "$$.bed";
my $num_to_check = 0;
open(TMPCOV, ">$covbed")
  or die "Cannot open temporary BED file $covbed for samtools: $!\n";
for my $chr (keys %$calls1){
  for my $pos (keys %{$calls1->{$chr}}){
    if(exists $subpar_calls2->{$chr} and exists $subpar_calls2->{$chr}->{$pos}){
      next; #exclude counting positions with sufficent coverage in one analysis, but not the other
    }
    else{
      $num_to_check++;
      print TMPCOV "$chr\t$pos\n";
    }
  }
}
close(TMPCOV);

print STDERR "Checking $num_to_check read depths for missing calls in $bam_file2...\n" unless $quiet;
my $remaining_calls1_homo = 0;
my $remaining_calls1_het = 0;
my $remaining_calls1_novel = 0;
my $remaining_calls1_novel_homo = 0;
open(SAM, "samtools mpileup -l $covbed -I $bam_file2 2>/dev/null |")
  or die "Cannot run samtools: $!\n";
while(<SAM>){
   my @B = split /\t/, $_;
   my $depth = @B > 1 ? $B[3] : 0;

   # The two methods may have used different patch versions of the same genome. Check if there was a no-call due to different reference bases.
   my $called_base = @B > 1 ? $B[2] : "N";
   if($depth and $called_base eq $calls1->{$B[0]}->{$B[1]}->[0]){
     next;
   }

   if($depth >= $calls2_min_depth){ # have good coverage in sample 2, but didn't make a SNP call like in sample 1
     if($calls1->{$B[0]}->{$B[1]}->[1] =~ /het/){
       $remaining_calls1_het++;
     }
     else{
       $remaining_calls1_homo++;
     }
     if($calls1->{$B[0]}->{$B[1]}->[3] eq "NA" or $calls1->{$B[0]}->{$B[1]}->[3] < 0.001){
       $remaining_calls1_novel++;
       $remaining_calls1_novel_homo++ if $calls1->{$B[0]}->{$B[1]}->[1] !~ /het/;
     }
     else{
       #print STDERR "Got method 1 only non-novel call: ", join("/", @{$calls1->{$B[0]}->{$B[1]}}), "\n";
     } 
   }
}
print OUT "Method 1 only\t",($remaining_calls1_homo+$remaining_calls1_het),"\n";
printf OUT "Method 1 only, het %%\t%.1f\n",$remaining_calls1_het/($remaining_calls1_homo+$remaining_calls1_het)*100;
printf OUT "Method 1 only, novel %%\t%.1f\n",$remaining_calls1_novel/($remaining_calls1_homo+$remaining_calls1_het)*100;
printf OUT "Method 1 only, novel, homo %%\t%.1f\n",$remaining_calls1_novel_homo/$remaining_calls1_novel*100;

$num_to_check = 0;
open(TMPCOV, ">$covbed")
  or die "Cannot open temporary BED file $covbed for samtools: $!\n";
for my $chr (keys %$calls2){
  for my $pos (keys %{$calls2->{$chr}}){
    if(exists $subpar_calls1->{$chr} and exists $subpar_calls1->{$chr}->{$pos}){
      next; #exclude counting positions with sufficent coverage in one analysis, but not the other
    }
    else{
      $num_to_check++;
      print TMPCOV "$chr\t$pos\n";
    }
  }
}
close(TMPCOV);

print STDERR "Checking $num_to_check read depths for missing calls in $bam_file1...\n" unless $quiet;
my $remaining_calls2_het = 0;
my $remaining_calls2_homo = 0;
my $remaining_calls2_novel = 0;
my $remaining_calls2_novel_homo = 0;
open(SAM, "samtools mpileup -l $covbed -I $bam_file1 2>/dev/null |")
  or die "Cannot run samtools: $!\n";
while(<SAM>){
   my @B = split /\t/, $_;
   my $depth = @B > 1 ? $B[3] : 0;

   # The two methods may have used different patch versions of the same genome. Check if there was a no-call due to different reference bases.
   my $called_base = @B > 1 ? $B[2] : "N";
   if($depth and $called_base eq $calls2->{$B[0]}->{$B[1]}->[0]){
     next;
   }

   if($depth >= $calls1_min_depth){ # have good coverage in sample 1, but didn't make a SNP call like in sample 2
     if($calls2->{$B[0]}->{$B[1]}->[1] =~ /het/){
       $remaining_calls2_het++;
     }
     else{
       $remaining_calls2_homo++;
     }
     if($calls2->{$B[0]}->{$B[1]}->[3] eq "NA" or $calls2->{$B[0]}->{$B[1]}->[3] < 0.001){
       $remaining_calls2_novel++;
       $remaining_calls2_novel_homo++ if $calls2->{$B[0]}->{$B[1]}->[1] !~ /het/;
     }
   }
}

print OUT "Method 2 only\t",($remaining_calls2_homo+$remaining_calls2_het),"\n";
printf OUT "Method 2 only, het %%\t%.1f\n",$remaining_calls2_het/($remaining_calls2_homo+$remaining_calls2_het)*100;
printf OUT "Method 2 only, novel %%\t%.1f\n",$remaining_calls2_novel/($remaining_calls2_homo+$remaining_calls2_het)*100;
printf OUT "Method 2 only, novel, homo %%\t%.1f\n",$remaining_calls2_novel_homo/$remaining_calls2_novel*100;
my $total_calls = $zygosities_same+$zygosities_diff+$bases_diff+$remaining_calls1_homo+$remaining_calls1_het+$remaining_calls2_homo+$remaining_calls2_het;
printf OUT "Total coding sequence genotype concordance%%\t%.1f\n", $zygosities_same/$total_calls*100;
printf OUT "Total coding sequence zygosity concordance%%\t%.1f\n", 100-$zygosities_diff/$total_calls*100;

unlink $covbed;

sub retrieve_calls{
  my ($hgvs_file, $reporting, $coding, $coverage_cutoff) = @_;
  my %calls;
  my %bad_coverage;
  open(HGVS, $hgvs_file)
    or die "Cannot open the first genotypes file $hgvs_file for reading: $!\n";
  <HGVS>; # header
  while(<HGVS>){
    my @F = split /\t/, $_;
    my $HGVS = $F[2];
    next unless $HGVS =~ />([ACGT])/; # only look at SNPs
    my $new_base = $1;
    $new_base =~ tr/ACGT/TGCA/ if $F[3] eq "-"; # rev comp cDNA HGVS
    my $chr = $F[4];
    my $pos = $F[5];
    my $zygosity = $F[6];
    my $coverage = $F[9];
    my $maf = $F[14];
    if($coverage_cutoff > $coverage or $F[16] ne ""){
      $bad_coverage{$chr} = {} unless exists $bad_coverage{$chr};
      $bad_coverage{$chr}->{$pos} = [$new_base, $zygosity, $coverage, $maf] unless exists $bad_coverage{$chr}->{$pos};
      next;
    }

    if(exists $reporting->{$chr}){ # in a region reported in the annotation, and it's protein coding?
      my $in_region = 0;
      my $arrayref = $reporting->{$chr};
      for(my $search_index = find_earliest_index($pos, $arrayref);$search_index <= $#{$arrayref}; $search_index++){
        my $interval = $arrayref->[$search_index];
        if($pos >= $interval->[0] and $pos <= $interval->[1]){
          $in_region = 1; last;
        }
        last if $pos < $interval->[0];
      }
      next unless $in_region;
      $in_region = 0;
      $arrayref = $coding->{$chr};
      for(my $search_index = find_earliest_index($pos, $arrayref);$search_index <= $#{$arrayref}; $search_index++){
        my $interval = $arrayref->[$search_index];
        if($pos >= $interval->[0] and $pos <= $interval->[1]){
          $in_region = 1; last;
        }
        last if $pos < $interval->[0];
      }
      next unless $in_region;
    }
  
    if(not exists $calls{$chr}){
      $calls{$chr} = {};
    }
    next if exists $calls{$chr}->{$pos};
    $calls{$chr}->{$pos} = [$new_base, $zygosity, $coverage, $maf];
  }
  close(HGVS);
  return (\%calls, \%bad_coverage);
}


sub find_earliest_index{
  # employs a binary search to find the smallest index that must be the starting point of a search of [start,end] elements sorted in an array by start
  my ($query, $array) = @_;

  return 0 if $query < $array->[0]->[0];

  my ($left, $right, $prevCenter) = (0, $#$array, -1);

  while(1) {
    my $center = int (($left + $right)/2);

    my $cmp = $query <=> $array->[$center]->[0] || ($center == 0 || $query != $array->[$center-1]->[0] ? 0 : -1);

    return $center if $cmp == 0;
    if ($center == $prevCenter) {
      return $left;
    }
    $right = $center if $cmp < 0;
    $left  = $center if $cmp > 0;
    $prevCenter = $center;
  }
}