Mercurial > repos > cpt > cpt_psm_recombine
comparison lib/CPT/Analysis/PAUSE.pm @ 1:97ef96676b48 draft
planemo upload commit 94b0cd1fff0826c6db3e7dc0c91c0c5a8be8bb0c
author | cpt |
---|---|
date | Mon, 05 Jun 2023 02:51:26 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
0:b18e8268bf4e | 1:97ef96676b48 |
---|---|
1 package CPT::Analysis::PAUSE; | |
2 | |
3 # ABSTRACT: Library for use in PAUSE analysis | |
4 use strict; | |
5 use warnings; | |
6 use Moose; | |
7 use List::Util qw(sum); | |
8 use Statistics::Descriptive; | |
9 | |
10 sub max ($$) { shift; $_[ $_[0] < $_[1] ] } | |
11 sub min ($$) { shift; $_[ $_[0] > $_[1] ] } | |
12 | |
13 sub derivative { | |
14 my ( $self, $data_ref ) = @_; | |
15 my @data = @{$data_ref}; | |
16 my @new_data; | |
17 foreach ( my $i = 0 ; $i < scalar(@data) - 1 ; $i++ ) { | |
18 $new_data[ $i + 1 ] = $data[ $i + 1 ] - $data[$i]; | |
19 } | |
20 return \@new_data; | |
21 } | |
22 | |
23 sub find_peaks { | |
24 my ( $self, %data ) = @_; | |
25 use IPC::Run3; | |
26 use File::Temp qw/tempfile/; | |
27 | |
28 # Store to CSV File | |
29 my @starts = @{ $data{data} }; | |
30 my ( $fh0, $filename0 ) = tempfile('galaxy.pause.XXXXXXX'); | |
31 printf $fh0 ( "%s,%s\n", 'position', 'count' ); | |
32 for ( my $i = 0 ; $i < scalar(@starts) ; $i++ ) { | |
33 printf $fh0 "%d,%d\n", $i, | |
34 ( defined $starts[$i] ? $starts[$i] : 0 ); | |
35 } | |
36 close($fh0); | |
37 | |
38 my ( $fh, $filename ) = tempfile('galaxy.pause.XXXXXXX'); | |
39 my @cmd = ( | |
40 'Rscript', $data{location_of_rscript_file}, | |
41 $filename0, $filename, $data{snr} | |
42 ); | |
43 my ( $in, $out, $err ); | |
44 run3 \@cmd, \$in, \$out, \$err; | |
45 | |
46 # Read in R data | |
47 my @values; | |
48 while (<$fh>) { | |
49 chomp; | |
50 push( @values, $_ ); | |
51 } | |
52 close($fh); | |
53 | |
54 unlink($filename0); | |
55 unlink($filename); | |
56 | |
57 return @values; | |
58 } | |
59 | |
60 sub smooth { | |
61 my ( $self, $data_ref ) = @_; | |
62 my @data = @{$data_ref}; | |
63 my @new_data; | |
64 my $length = scalar @data; | |
65 foreach ( my $i = 0 ; $i < $length ; $i++ ) { | |
66 my $avg = | |
67 sum( @data[ $i - 20 .. $i - 1, $i + 1 .. $i + 20 ] ) / 40; | |
68 $new_data[$i] = $avg; | |
69 } | |
70 return \@new_data; | |
71 } | |
72 | |
73 sub histogram { | |
74 my ( $self, %data ) = @_; | |
75 | |
76 my @coverage = @{ $data{data} }; | |
77 my @return_coverage; | |
78 for ( my $i = 0 ; $i < scalar(@coverage) ; $i++ ) { | |
79 my $size = $coverage[$i]; | |
80 unless ($size) { $size = 0 } | |
81 $return_coverage[$i] = [ $i, $size, "*" x $size ]; | |
82 } | |
83 my %results = ( | |
84 'Sheet1' => { | |
85 headers => [qw(Base Count Plot)], | |
86 data => \@return_coverage, | |
87 } | |
88 ); | |
89 return %results; | |
90 } | |
91 | |
92 sub getCoverageDensity { | |
93 my ( $self, %data ) = @_; | |
94 | |
95 # Load the sam file | |
96 my $sam = Bio::DB::Sam->new( | |
97 -bam => $data{bam}, | |
98 -fasta => $data{genome}, | |
99 -autoindex => 1, | |
100 ); | |
101 | |
102 # Get all alignments to our indicated FASTA file | |
103 my @alignments = $sam->get_features_by_location( | |
104 -seq_id => $data{fasta_id}, | |
105 -start => 1, | |
106 -end => $data{fasta_length} | |
107 ); | |
108 | |
109 # Set up some variables | |
110 my $coverage_density_max_value = 0; | |
111 my ( @coverage_density, @read_starts, @read_ends ); | |
112 | |
113 # including some for statistics | |
114 my $stat_start = Statistics::Descriptive::Sparse->new(); | |
115 my $stat_end = Statistics::Descriptive::Sparse->new(); | |
116 | |
117 # Looping over alignments | |
118 for my $a (@alignments) { | |
119 my $start = $a->start; | |
120 my $end = $a->end; | |
121 | |
122 # Increment the number of reads starting there | |
123 $read_starts[$start]++; | |
124 $read_ends[$end]++; | |
125 | |
126 # And increment the coverage density | |
127 foreach ( $start .. $end ) { | |
128 $coverage_density[$_]++; | |
129 if ( $coverage_density[$_] > | |
130 $coverage_density_max_value ) | |
131 { | |
132 $coverage_density_max_value = | |
133 $coverage_density[$_]; | |
134 } | |
135 } | |
136 } | |
137 my @start_data_for_stats; | |
138 my @end_data_for_stats; | |
139 for ( my $i = 0 ; $i < $data{fasta_length} ; $i++ ) { | |
140 if ( $read_starts[$i] ) { | |
141 push( @start_data_for_stats, $read_starts[$i] ); | |
142 } | |
143 if ( $read_ends[$i] ) { | |
144 push( @end_data_for_stats, $read_ends[$i] ); | |
145 } | |
146 } | |
147 $stat_start->add_data(@start_data_for_stats); | |
148 $stat_end->add_data(@end_data_for_stats); | |
149 | |
150 # Lots of data to return | |
151 use CPT::Analysis::PAUSE::ParsedSam; | |
152 my $psam = CPT::Analysis::PAUSE::ParsedSam->new( | |
153 coverage_density => \@coverage_density, | |
154 read_starts => \@read_starts, | |
155 read_ends => \@read_ends, | |
156 max => $coverage_density_max_value, | |
157 stats_start_max => $stat_start->max(), | |
158 stats_end_max => $stat_end->max(), | |
159 stats_start_mean => $stat_start->mean(), | |
160 stats_end_mean => $stat_end->mean(), | |
161 stats_start_standard_deviation => | |
162 $stat_start->standard_deviation(), | |
163 stats_end_standard_deviation => $stat_end->standard_deviation(), | |
164 ); | |
165 return $psam; | |
166 } | |
167 | |
168 no Moose; | |
169 1; | |
170 | |
171 __END__ | |
172 | |
173 =pod | |
174 | |
175 =encoding UTF-8 | |
176 | |
177 =head1 NAME | |
178 | |
179 CPT::Analysis::PAUSE - Library for use in PAUSE analysis | |
180 | |
181 =head1 VERSION | |
182 | |
183 version 1.96 | |
184 | |
185 =head1 AUTHOR | |
186 | |
187 Eric Rasche <rasche.eric@yandex.ru> | |
188 | |
189 =head1 COPYRIGHT AND LICENSE | |
190 | |
191 This software is Copyright (c) 2014 by Eric Rasche. | |
192 | |
193 This is free software, licensed under: | |
194 | |
195 The GNU General Public License, Version 3, June 2007 | |
196 | |
197 =cut |