#!/usr/bin/env ruby # # bio-gngm # # Created by Dan MacLean (TSL) on 2011-12-07. # Copyright (c) . All rights reserved. ################################################### require 'rubygems' require 'rinruby' require 'bio-samtools' require 'bio/db/pileup' require 'bio/db/vcf' require 'pp' =begin Simple class representing a file of Fasta format sequences and each ones length =end class Bio::DB::FastaLengthDB require 'bio' def initialize(args) @file = args[:file] @seqs = {} file = Bio::FastaFormat.open(@file) file.each do |entry| @seqs[entry.entry_id] = entry.length end def each @seqs.keys.sort.each do |k| yield k, @seqs[k] end end end end =begin Extends the methods of the Bio::DB::Pileup class in bio-samtools. A pileup object represents the SAMtools pileup format at http://samtools.sourceforge.net/pileup.shtml. These extension methods are used by the Bio::Util::Gngm object internally and are not exposed to the user of the Bio::Util::Gngm object through that. =end class Bio::DB::Pileup #attributes set by call to Bio::DB::Pileup#discordant_chastity attr_accessor :top_non_ref_count, :second_non_ref_count, :third_non_ref_count #calculates the discordant chastity statistic as defined in Austin et al (2011) http://bar.utoronto.ca/ngm/description.html and http://onlinelibrary.wiley.com/doi/10.1111/j.1365-313X.2011.04619.x/abstract;jsessionid=F73E2DA628523B26205297CEE95526DA.d02t04 #Austin _et_ _al_ (2011) *Next-generation* *mapping* *of* *Arabidopsis* *genes* _Plant_ _Journal_ *67*(4):7125-725 # #Briefly, # The statistic measures the degree of difference between the SNP and the expected reference base. # Using the mapping information comprising a SNP, the most frequent base that is not the reference base # is compared to the next most common base after it. # (from http://bar.utoronto.ca/ngm/description.html ) def discordant_chastity arr = self.non_refs.to_a.sort {|a,b| b.last <=> a.last } @top_non_ref_count, @second_non_ref_count, @third_non_ref_count = arr.collect {|c| c.last} case when self.non_ref_count == 0 then 0.0 when @top_non_ref_count == @coverage then 1.0 when @second_non_ref_count > 0 then @top_non_ref_count.to_f / (@top_non_ref_count + @second_non_ref_count).to_f else @top_non_ref_count.to_f / @coverage.to_f end end #returns true if self is a SNP with minimum coverage depth of +:min_depth+ and minimum non-reference bases of +:min_non_ref_count+ #returns false for every position where the reference base is N or n if +:ignore_reference_n+ is set to true # #Options and Defaults: #- :min_depth => 2 #- :min_non_ref_count => 2 #- :ignore_reference_n => false # #Example # pileup.is_snp?(:min_depth => 5, :min_non_ref_count => 2) # pileup.is_snp?(:min_depth => 5, :min_non_ref_count => 1, :ignore_reference_n => true) def is_snp?(opts) return false if self.ref_base == '*' #return false unless is_ct return false if opts[:ignore_reference_n] and self.ref_base == "N" or self.ref_base == "n" return true if self.coverage >= opts[:min_depth] and self.non_ref_count >= opts[:min_non_ref_count] false end end #Extends the methods of the Bio::DB::Vcf class in bio-samtools. A Vcf object represents the VCF format described at #http://www.1000genomes.org/node/101 . The Bio::DB::Vcf object returns all information in the VCF line, but the implementation here acts as if #there is only possibly one variant at each position and ignores positions at which there may be multiple variants. Vcf format is only used when #the Bio::Util::Gngm object requests information about indels using SAMtools mpileup method. class Bio::DB::Vcf #returns true if the +alt+ column of the Vcf is not *.* # #Examples # #vcf record = 20 14370 rs6054257 G A 29 PASS ... # vcf.variant? #=> true #vcf record = 20 1230237 . T . 47 PASS ... # vcf.variant? #=> false def variant? not self.alt == "." rescue false end #Return a short string representing chromosome, position, reference sequence, alt sequence(s) and the info string of the Vcf object. def to_s "#{self.chrom} #{self.pos} #{self.ref} #{self.alt} #{self.info}" end #The depth of reads actually used in the genotype call by Vcftools. The sum of the DP4 attribute. Returns 0.0 if no value is calculated. def used_depth self.info["DP4"].split(",").inject {|sum,n| sum.to_f + n.to_f} rescue 0.0 end #List of alternate alleles at this locus, obtained by splitting the vcf.alt attribute string on commas # #Example #vcf.alt = "ACT,TCA" # vcf.alternatives = ["ACT", "TCA"] #vcf.alt = "T" # vcf.alternatives = ["T"] def alternatives self.alt.split(",") rescue [] end ##Returns the depth of reads containing the non reference allele. IE the sum of the last two figures in the DP4 attribute. def non_ref_allele_count self.info["DP4"].split(",")[2..3].inject {|sum,n| sum.to_f + n.to_f } rescue 0.0 end #Returns the non-reference allele frequency based on depth of reads used for the genotype call, # #IE # vcf.non_ref_allele_count / vcf.used_depth def non_ref_allele_freq self.non_ref_allele_count / self.used_depth end #Returns the mean Mapping Quality from the reads over this position as defined by the Vcf MQ attribute. def mq self.info["MQ"].to_f rescue 0.0 end ##Returns the genotype quality score from the sample data (as defined by the Vcf GQ attribute) for the first sample in the Vcf only. def gq self.samples["1"]["GQ"].to_f rescue 0.0 end #Returns the phred scaled likelihood of the first non-reference allele (as defined by the Vcf PL attribute) for the first sample in the Vcf only. def pl self.samples["1"]["PL"].split(",")[1].to_f rescue 0.0 end #Returns true if only one variant allele is recorded. Loci with more than one allele are too complicated for now, so are discarded... #def has_just_one_variant? # self.alternatives.length == 1 and self.variant? #end #Returns true if the position passes criteria # #Options and Defaults: #- :min_depth => 2 #- :min_non_ref_count => 2 #- :mapping_quality => 10 # #Example # vcf.pass_quality?(:min_depth => 5, :min_non_ref_count => 2, :mapping_quality => 25, :min_snp_quality => 20) def pass_quality?(options) (self.used_depth >= options[:min_depth] and self.mq >= options[:mapping_quality] and self.non_ref_allele_count >= options[:min_non_ref_count] and self.qual >= options[:min_snp_quality]) end #returns true if ref col has same length as all alternatives and position variant passes quality def is_mnp?(options) return true if self.alternatives.all? {|x| x.length == self.ref.length} and self.pass_quality?(options) false end ##returns true if ref col has length of 1 and is_mnp? def is_snp?(options) return true if self.is_mnp?(options) and self.ref.length == 1 false end #Returns true if ref col is different in length from any of the entries in alt column def is_indel?(options) return true if self.variant? and self.alternatives.any? {|x| x.length != self.ref.length} and self.pass_quality?(options) false end end module Bio class Util =begin A Bio::Util::Gngm object represents a single region on a reference genome that is to be examined using the NGM technique described in Austin et al (2011) http://bar.utoronto.ca/ngm/description.html and http://onlinelibrary.wiley.com/doi/10.1111/j.1365-313X.2011.04619.x/abstract;jsessionid=F73E2DA628523B26205297CEE95526DA.d02t04 Austin _et_ _al_ (2011) *Next-generation* *mapping* *of* *Arabidopsis* *genes* _Plant_ _Journal_ *67*(4):7125-725 . Bio::Util::Gngm provides methods for finding SNPs, small INDELS and larger INDELS, creating histograms of polymorphism frequency, creating and clustering density curves, creating signal plots and finding peaks. The ratio of reference-agreeing and reference-differing reads can be specified. == Background The basic concept of the technique is that density curves of polymorphism frequency across the region of interest are plotted and analysed. Each curve is called a thread, as it represents a polymorphism that was called with a statistic within a certain user-specified range, eg if a SNP was called with 50% non-reference bases from sequence reads (say all A), and 50% reference reads (all T) then a discordant chastity statistic (ChD) of 0.5 would be calculated and assigned to that SNP. Depending on the width and slide of the windows the user had specified, the frequency of SNPs with ChD in the specified range would be drawn in the same density curve. In the figure below each different coloured curve represents the frequency of SNPs with similar ChD. link:images/threads.png Each of these density curves is called a thread. Threads are clustered into groups called bands and the bands containing the expected and control polymorphisms extracted. In the figure below, the control band is 0.5, the expected mutation in 1.0. Typically and in the Austin et al (2011) description of NGM the control band is the heterophasic band that represents natural variation, the thing taken to be the baseline. For a simple SNP, numerically the discordant chastity is expected to be 0.5. Conversely the expected band is the homophasic band that represents the selected for SNP region. Normally the discordant chastity is expected to be 1.0. link:images/bands.png The points where the signal from the control and expected band converge most is a likely candidate region for the causative mutation, so here at about the 1.6 millionth nucleotide. link:images/signal.png == Example require 'bio-gngm' g = Bio::Util::Gngm.new(:file => "aln.sorted.bam", :format => :bam, :fasta => "reference.fasta", :start => 100, :stop => 200, :write_pileup => "my_pileup_file.pileup", :write_vcf => "my_vcf_file.vcf", :ignore_file => "my_known_snps.txt" :samtools => { :q => 20, :Q => 50 }, :min_non_ref_freq => 0.5, :min_non_ref => 3, :start => 1, :stop => 100000, :chromosome => "Chr1", :variant_call => { :indels => false, :min_depth => 6, :max_depth => 250, :mapping_quality => 20.0, :min_non_ref_count => 2, :ignore_reference_n => true, :min_snp_quality => 20, :min_consensus_quality => 20, :substitutions => ["C:T","G:A"] } ) g.snp_positions g.collect_threads(:start => 0.2, :stop => 1.0, :slide => 0.01, :size => 0.1 ) [0.25, 0.5, 1.0].each do |kernel_adjust| # loop through different kernel values [4, 9, 11].each do | k | # loop through different cluster numbers #cluster g.calculate_clusters(:k => k, :adjust => kernel_adjust, :control_chd => 0.7, :expected_chd => 0.5) #draw thread and bands filename = "#{name}_#{k}_#{kernel_adjust}_all_threads.png" g.draw_threads(filename) filename = "#{name}_#{k}_#{kernel_adjust}_clustered_bands.png" g.draw_bands(filename, :add_lines => [100,30000,675432]) #draw signal filename = "#{name}_#{k}_#{kernel_adjust}_signal.png" g.draw_signal(filename) #auto-guess peaks filename = "#{name}_#{k}_#{kernel_adjust}_peaks.png" g.draw_peaks(filename) end end g.close #close BAM file == Polymorphisms and statistics Bio::Util::Gngm will allow you to look for polymorphisms that are SNPs, INDELS (as insertions uniquely, deletions uniquely or both) and longer insertions or deletions based on the insert size on paired-end read alignments. Each has a different statistic attached to it. === SNPs Simple Single Nucleotide Polymorphisms are called and its ChD statistic calculated as described in Austin et al (2011). === Short INDELS These are called via SAMtools/BCFtools so are limited to the INDELs that can be called that way. The implementation at the moment only considers positions with one INDEL, sites with more than one potential INDEL (ie multiple alleles) are disregarded as a position at all. See the Bio::DB::Vcf extensions in this package for a description of what constitutes an INDEL. The Vcf attribute Bio::DB::Vcf#non_ref_allele_freq is used as the statistic in this case. === Insertion Size Paired-end alignments have an expected distance between the paired reads (called insert size, or isize). Groups of reads in one position with larger or smaller than expected isize can indicate large deletions or insertions. Due to the details of read preparation the actual isize varies around a mean value with an expected proportion of 50% of reads having isize above the mean, and 50% below. To create density curves of insertion size frequency a moves along the window of user-specified size is moved along the reference genome in user-specified steps and all alignments in that window are examined. The Bio::DB::Sam#isize attribute is inspected for all alignments passing user-specified quality and the proportion of reads in that window that have an insert size > the expected insert size is used as the statistic in this case. Proportions approaching 1 indicate that the sequenced organism has a deletion in that section relative to the reference. Proportions approaching 0 indicate an insertion in that section relative to the reference. Proportions around 0.5 indicate random variation of insert size, IE no INDEL. Seems to be a good idea to keep the window size similar to the read + isize. Useful in conjunction with assessing unmapped mates. === Unmapped Mate Pairs / Paired Ends. Paired-end alignments where one mate finds a mapping but the other doesnt, can indicate an insertion/deletion larger than the insert size of the reads used (IE one read disappeared into the deleted section). This method uses a statistic based on proportion of mapped/unmapped reads in a window. Proportions of reads that are mapped but the mate is unmapped should be about 0.5 in a window over an insertion/deletion (since the reads can go in either direction..). With no insertion deletion, the proportion should be closer to 0. == Input types A sorted BAM file is used as the source of alignments. Pileup is not used nor likely to be as it is a deprecated function within SAMtools. With the BAM file you will need the reference FASTA and the BAM index (.bai). == Workflow 1. Create Bio::Util::Gngm object for a specific region in the reference genome 2. Polymorphisms are found 3. Density curves (threads) are calculated 4. Clustering density threads into bands is done 5. Signal is compared between band of interest and control 6. Figures are printed == Prerequisites - Ruby 1.9.3 or greater (if you have an earlier version, try RVM for installing different versions of Ruby alongside your system install and switching nicely between them) - R 2.11.1 or greater The following ruby-gems are required - rinruby >= 2.0.2 - bio-samtools >= 0.5.0 The following R packages are required - ggplot2 - peaks == Acknowledgements Thanks very much indeed to Ryan Austin, who invented NGM in the first place and was very forthcoming with R code, around which this implementation is based. == Using bio-gngm require 'bio-gngm' == API =end class Gngm attr_accessor :file #Ruby 1.9.3 has a rounding error in the Range#step function such that some decimal places are rounded off to 0.00000000000000...1 above their place. So this constant is used to identify #windows within a short distance and prevent any rounding errors. Hopefully I should be able to remove this in later versions. ERROR_MARGIN = 0.000001 public #Returns a new Bio::Util::Gngm object. # # g = Bio::Util::Gngm.new(:file => "aln.sort.bam", # :format => :bam, # :samtools => {:q => 20, :Q => 50}, # :fasta => "reference.fa" # :start => 100, # :stop => 200, # :write_pileup => "my_pileup_file.pileup", # :write_vcf => "my_vcf_file.vcf", # :ignore_file => "my_known_snps.txt" # # ) # #Required parameters and defaults: #- :file => nil -the path to the bam file containing the alignments, a .bai index must be present. A pileup file, or tab-delimited text file can be used. #- :format => :bam -either :bam, :pileup, :txt (pileup expected to be 10 col format from samtools -vcf) #- :chromosome => "nil" -sequence id to look at #- :start => nil -start position on that sequence #- :stop => nil -stop position on that sequence #- :fasta => nil -the path to the FASTA formatted reference sequence #- :write_pileup => false -the path to a file. SNPs will be written in pileup to this file (indels not output) #- :write_vcf => false -the path to a file. SNPs will be written in VCF to this file (indels not output) #- :ignore_file => false -file of SNPs in format "reference sequence id \t position \t mapping line nucleotide identity \t reference line nucleotide identity". All SNPs in this file will be ignored #- :samtools => {:q => 20, :Q => 50} -options for samtools, see bio-samtools documentation for further details. #Optional parameters and defaults: # #Most of these are parameters for specific methods and can be over-ridden when particular methods are called #- :variant_call => {:indels => false, #- :min_depth => 2, #- :max_depth => 10000000, #- :min_snp_quality => 20, #- :mapping_quality => 10.0, #- :min_non_ref_count => 2, #- :ignore_reference_n => true, #- :min_consensus_quality => 20, #- :min_snp_quality => 20 }. # - For Pileup files from old samtools pileup -vcf :min_consensus_quality can be applied #- :threads => {:start => 0.2, :stop => 1.0, :slide => 0.01, :size => 0.1 } -options for thread windows #- :insert_size_opts => {:ref_window_size => 200, :ref_window_slide => 50, :isize => 150} -options for insert size calculations #- :histo_bin_width => 250000 -bin width for histograms of SNP frequency #- :graphics => {:width => 1000, :height => 500, :draw_legend => false, :add_boxes => nil} -graphics output options, +:draw_legend+ draws a legend plot for band figures only #- :peaks => {:sigma => 3.0, :threshold => 10.0, :background => false, :iterations => 13, :markov => false, :window => 3, :range => 10000} -parameters for automated peak calling, parameters relate to R package Peaks. +:range+ is the width of the box to draw on the peak plot def initialize(options) @file = nil @snp_positions = nil @threads = nil @densities = nil @clusters = nil @control_band = nil @expected_band = nil @signal = nil @peak_indices = nil @peak_y_values = nil @density_max_y = nil #the maximum y value needed to plot the entire set density plots of threads and maintain a consistent scale for plots @colours = %w[#A6CEE3 #1F78B4 #B2DF8A #33A02C #FB9A99 #E31A1C #FDBF6F #FF7F00 #CAB2D6 #6A3D9A #FFFF99 #B15928] @thread_colours = {} @known_variants = nil #a list of variants to keep track of @opts = { :file => nil, :format => :bam, :fasta => nil, :samtools => {:q => 20, :Q => 50}, :indels => false, :write_pileup => false, :write_vcf => false, :ignore_file => false, :insert_size_opts => {:ref_window_size => 200, :ref_window_slide => 50, :isize => 150}, :variant_call => { :indels => false, :min_depth => 2, :max_depth => 10000000, :mapping_quality => 10.0, :min_non_ref_count => 2, :ignore_reference_n => true, :shore_map => false, :snp_file => :false, :min_consensus_quality => 20, :min_snp_quality => 20}, ## some options are designed to be equivalent to vcfutils.pl from bvftools options when using vcf ##:min_depth (-d) ##:max_depth (-D) ##:mapping_quality (-Q) minimum RMS mappinq quality for SNPs (mq in info fields) ##:min_non_ref_count (-a) minimum num of alt bases ... the sum of the last two numbers in DP4 in info fields ##doesnt do anything with window filtering or pv values... :histo_bin_width => 250000, :graphics => {:width => 1000, :height => 500, :draw_legend => false, :add_boxes => nil}, :adjust => 1, :control_chd => 0.5, :expected_chd => 1.0, :threads => {:start => 0.2, :stop => 1.0, :slide => 0.01, :size => 0.1 }, :peaks => {:sigma => 3.0, :threshold => 10.0, :background => false, :iterations => 13, :markov => false, :window => 3, :range => 10000} ##range is the width of the box to draw on the peak plot } @opts.merge!(options) @opts[:samtools][:r] = "#{options[:chromosome]}:#{options[:start]}-#{options[:stop]}" @pileup_outfile, @vcf_outfile = nil,nil if @opts[:variant_call][:indels] and (@opts[:write_pileup] or @opts[:write_vcf]) $stderr.puts "Cannot yet output VCF/Pileup when generating INDELs. Turning output off." @opts[:write_pileup] = false @opts[:write_vcf] = false end if @opts[:write_pileup] @pileup_outfile = File.open(@opts[:write_pileup], "w") end if @opts[:write_vcf] @vcf_outfile = File.open(@opts[:write_vcf], "w") end @known_snps = Hash.new if @opts[:ignore_file] File.open(@opts[:ignore_file], "r").each do |line| col = line.chomp.split(/\t/) if @known_snps[col[0]] @known_snps[col[0]][col[1].to_i] = 1 else @known_snps[col[0]] = Hash.new @known_snps[col[0]][col[1].to_i] = 1 end end end open_file end private #opens the file def open_file case @opts[:format] when :bam then open_bam when :pileup, :text then open_text end end private #calls Bio::DB::Sam.open def open_bam @file = Bio::DB::Sam.new(:bam => @opts[:file], :fasta => @opts[:fasta] ) @file.open end def open_text @file = File.open(@opts[:file], "r") end public #for BAM files calls Bio::DB::Sam#close to close the connections to input files safely def close case @opts[:format] when :bam then @file.close end end public #Returns array of arrays [[position, statistic]] for polymorphisms passing filters in +optsa+ #Default options are those in the +:variant_call+ global options hash which can be over ridden in the method call # #Options and defaults: #- :indels => false -call small insertions AND deletions instead of simple SNPs #- :deletions_only => false -call just deletions instead of simple SNPs #- :insertions_only => false -call small insertions instead of simple SNPs #- :min_depth => 2 -minimum quality passing depth of coverage at a position for a SNP call #- :max_depth => 10000000 -maximum quality passing depth of coverage at a position for a SNP call #- :mapping_quality => 10.0 -minimum mapping quality required for a read to be used in depth calculation #- :min_non_ref_count => 2 -minimum number of reads not matching the reference for SNP to be called #- :ignore_reference_n => true -ignore positions where the reference is N or n #When INDEL calling only one of :indels should be used. If +false+, SNPs are called. # #calculates or returns the value of the instance variable @snp_positions. Only gets positions the first time it is called, in subsequent calls pre-computed positions and statistics are returned, so changing parameters has no effect. def snp_positions(optsa={}) opts = @opts[:variant_call].merge(optsa) return @snp_positions if @snp_positions case @opts[:format] when :bam then get_snp_positions_from_bam(opts) when :text then get_snp_positions_from_text(opts) when :pileup then get_snp_positions_from_pileup(opts) end end ##allows the user to assign SNP positions def snp_positions=(arr) @snp_positions = arr end def is_allowed_substitution?(ref,alt,opts) if opts[:substitutions].instance_of?(Array) return false unless opts[:substitutions].include?("#{ref}:#{alt}") end true end private #Calls SNP/short INDEL positions from a BAM file and the appropriate statistic according to quality criteria passed by Bio::Util::Gngm#snp_positions. #Sets @snp_positions def get_snp_positions_from_bam(options={}) opts = @opts[:variant_call].merge(options) arr = [] ##when we are calling mpileup_plus we need to add :g to the samtools options #alw @opts[:samtools][:g] = true if opts[:indels] if not @opts[:samtools][:g] @file.mpileup(@opts[:samtools]) do |pileup| if pileup.is_snp?(opts) and is_allowed_substitution?(pileup.ref_base, pileup.consensus,opts) and not @known_snps[pileup.ref_name][pileup.pos] arr << [pileup.pos, pileup.discordant_chastity] write(pileup) end end else @file.mpileup_plus(@opts[:samtools]) do |vcf| next if not vcf.variant? ##we dont care about the calls for reference agreeing positions next if (opts[:ignore_reference_n] and vcf.ref =~ /N/i) ##indel use returns the vcf allele_frequency, not the ChDs (because calculating it is a mess... ) if opts[:indels] arr << [vcf.pos, vcf.non_ref_allele_freq] if vcf.is_indel?(opts) and is_allowed_substitution?(vcf.ref, vcf.alt,opts) and not @known_snps[vcf.ref][vcf.pos] else arr << [vcf.pos, vcf.non_ref_allele_freq] if vcf.is_snp?(opts) and is_allowed_substitution?(vcf.ref, vcf.alt,opts) and not @known_snps[vcf.ref][vcf.pos] end end end @snp_positions = arr arr end #this does not filter snps, other than to check they are in the right region and are allowed substitutions.. no qual control, assumed to be done prior #text file is of format chr\tpos\tref\talt\tfreq\n def get_snp_positions_from_text(options={}) arr = [] opts = @opts[:variant_call].merge(options) @file.each do |line| chr,pos,ref,alt,freq = line.chomp.split("\t") pos = pos.to_i freq = freq.to_f next unless chr == @opts[:chromosome] and pos >= @opts[:start] and pos <= @opts[:stop] and is_allowed_substitution?(ref,alt,opts) and not @known_snps[chr][pos] arr << [pos, freq] end @snp_positions = arr end private def get_snp_positions_from_pileup(options={}) arr = [] opts = @opts[:variant_call].merge(options) @file.each do |line| pileup = Bio::DB::Pileup.new(line) if pileup.ref_name != @opts[:chromosome] or pileup.pos < @opts[:start] or pileup.pos > @opts[:stop] next end #old fashioned 10 col pileup format has extra fields we can use if needed if pileup.is_snp?(opts) and not pileup.consensus_quality.nil? and not pileup.snp_quality.nil? and not @known_snps[pileup.ref_name][pileup.pos] write(pileup) arr << [pileup.pos, pileup.discordant_chastity] if pileup.consensus_quality > opts[:min_consensus_quality] and pileup.snp_quality > opts[:min_snp_quality] and is_allowed_substitution?(pileup.ref_base, pileup.consensus,opts) end end @snp_positions = arr end private #writes out pileup/vcf files of SNPs that were used def write(obj) if @opts[:write_pileup] @pileup_outfile.puts(obj.to_s) end if @opts[:write_vcf] @vcf_outfile.puts(obj.to_vcf) end end private #Gets the insert size for each alignment in the BAM positions from a BAM file according to quality criteria passed by Bio::Util::Gngm#get_insert_size_frequency. def get_insert_size_frequency_from_bam(opts={}) reference_window_size,reference_window_slide = opts[:ref_window_size], opts[:ref_window_slide] arr = [] @opts[:samtools][:r] =~ /(.*):(.*)-(.*)/ chr,rstart,rstop = $1.to_s,$2.to_i,$3.to_i (rstart..rstop).step(reference_window_slide) do |win_start| win_tot = 0.0 win_over_isize = 0.0 @file.fetch(chr, win_start, win_start + reference_window_size).each do |alignment| next if not alignment_passes(alignment) win_tot = win_tot + 1 win_over_isize = win_over_isize + 1 if alignment.isize.abs > opts[:isize] end prop = win_over_isize / win_tot arr << [win_start, prop] end @snp_positions = arr end #Gets the proportion of reads with unmapped mates in a window def get_unmapped_mate_frequency_from_bam(opts={}) reference_window_size,reference_window_slide = opts[:ref_window_size], opts[:ref_window_slide] arr = [] @opts[:samtools][:r] =~ /(.*):(.*)-(.*)/ chr,rstart,rstop = $1.to_s,$2.to_i,$3.to_i (rstart..rstop).step(reference_window_slide) do |win_start| #puts "__________________#{win_start}____________________" win_tot = 0.0 win_mates_unmapped = 0.0 @file.fetch(chr, win_start, win_start + reference_window_size).each do |alignment| next if (alignment.failed_quality) # or @opts[:samtools][:q] <= alignment.mapq or not alignment.is_paired) win_tot = win_tot + 1 win_mates_unmapped = win_mates_unmapped + 1 if alignment.mate_unmapped end #puts "win tot #{win_tot}" #puts "win mates #{win_mates_unmapped}" prop = win_mates_unmapped / win_tot #puts "prop #{prop}" arr << [win_start, prop] end @snp_positions = arr end private #Returns true if the passed Bio::DB::Sam passes the quality criteria def alignment_passes(aln) not aln.failed_quality && @opts[:samtools][:q] <= aln.mapq && aln.is_paired and not aln.mate_unmapped end public #Returns array of arrays [[window start position, proportion of alignments > insert size]]. #Does this by taking successive windows across reference and collects the proportion of the reads in that window #that have an insert size > the expected insert size. Proportions approaching 1 indicate that the sequenced organism has a deletion in that section, proportions approaching 0 indicate an insertion in that section, proportions around 0.5 indicate random variation of insert size, IE no indel. # #Each section should be approximately the size of the insertion you expect to find and should increment in as small steps as possible. # #Options and defaults: #- :ref_window_size => 200 width of window in which to calculate proportions #- :ref_window_slide => 50 number of bases to move window in each step #- :isize => 150 expected insert size # #Sets the instance variable @snp_positions. Only gets positions the first time it is called, in subsequent calls pre-computed positions and statistics are returned, so changing parameters has no effect def get_insert_size_frequency(options={}) opts = @opts[:insert_size_opts].merge(options) return @snp_positions if @snp_positions case when @file.instance_of?(Bio::DB::Sam) then get_insert_size_frequency_from_bam(opts) end end #Returns array of arrays [[window start position, proportion of reads with unmapped mates]]. #Does this by taking successive windows across reference and counting the reads with unmapped mates #Proportions approaching 0.5 indicate that the sequenced organism has an insertion in that section, proportions approaching 0 indicate nothing different in that section. # #Each section should be approximately the size of the insertion you expect to find and should increment in as small steps as possible. # #Options and defaults: #- :ref_window_size => 200 width of window in which to calculate proportions #- :ref_window_slide => 50 number of bases to move window in each step # #Sets the instance variable @snp_positions. Only gets positions the first time it is called, in subsequent calls pre-computed positions and statistics are returned, so changing parameters has no effect def get_unmapped_mate_frequency(options={}) opts = @opts[:insert_size_opts].merge(options) return @snp_positions if @snp_positions case when @file.instance_of?(Bio::DB::Sam) then get_unmapped_mate_frequency_from_bam(opts) end end public #Draws a histogram of polymorphism frequencies across the reference genome section defined in Bio::Util::Gngm#initialize with bin width +bin_width+ and writes it to a PNG file +file+ def frequency_histogram(file="myfile.png", bin_width=@opts[:histo_bin_width], opts=@opts[:graphics]) posns = self.snp_positions.collect {|a| a.first} r = new_r r.eval "suppressMessages ( library(ggplot2) )" #setup R environment... r.posns = posns r.eval "data = data.frame(position=posns)" r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" graph_cmd = "qplot(position,data=data, geom='histogram', binwidth = #{bin_width}, alpha=I(1/3), main='#{file}', color='red')" r.eval(graph_cmd) r.eval "dev.off()" r.quit end #Returns contents of @threads, an array of arrays [[window 1, snp position 1, snp position 2 ... snp position n],[window 2, snp position 1, snp position 2 ... snp position n] ]. #If @threads is nil (because snps have not yet been gathered into threads) the Bio::Util::Gngm#collect_threads method is called and @threads is set before returning # #Options and defaults: #- :start => 0.2 -first window #- :stop => 1.0 -last window #- :slide => 0.01 -distance between windows #- :size => 0.1 -window width public def threads(opts=@opts[:threads]) @threads ||= collect_threads(opts) end public #Resets contents of instance variable @threads and returns an array of arrays [[window 1, snp position 1, snp position 2 ... snp position n],[window 2, snp position 1, snp position 2 ... snp position n] ]. #Always sets @threads regardless of whether it contains anything or not so is useful for trying out different window sizes etc # #Options and defaults: #- :start => 0.2 -first window #- :stop => 1.0 -last window #- :slide => 0.01 -distance between windows #- :size => 0.1 -window width def collect_threads(options={}) opts = @opts[:threads].merge(options) opts[:slide] = 0.000001 if opts[:slide] < 0.000001 ##to allow for the rounding error in the step function... raise RuntimeError, "snp positions have not been calculated yet" if not @snp_positions start,stop,slide,size = opts[:start].to_f, opts[:stop].to_f, opts[:slide].to_f, opts[:size].to_f arr = [] (start..stop).step(slide) do |win| arr << [win, @snp_positions.select {|x| x.last >= win and x.last < win + size }.collect {|y| y.first} ] end @threads = arr end private #Returns the value of @density_max_y or if nil, calls Bio::Util::Gngm#get_density_max_y to work out the maximum y axis value for plots #Might not work properly as seems to call non-existent method... def density_max_y @density_max_y ||= get_density_max_y end private def calculate_density_max_y mx = 0.0 self.densities.each do |x| m = x[2].max mx = m if m > mx end @density_max_y = mx end public #Draws the threads in a single PNG file +file+ # #Options and defaults #- :draw_legend => nil -if a filename is provided a legend will be drawn in a second plot #- :width => 1000 -width of the PNG in pixels #- :height => 500 -height of the PNG in pixels def draw_threads(file="myfile.png", options={}) opts = @opts[:graphics].merge(options) #uses R's standard plot functions.. needed because ggplot can die unexpectedly... raise RuntimeError, "Can't draw threads until clustering is done" unless @clusters r = new_r r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" plot_open = false self.densities.each do |t| r.curr_win = t.last r.dx = t[1] r.dy = t[2] if plot_open r.eval "lines(dx,dy, col=\"#{@thread_colours[t.first]}\", xlab='position', ylab='density')" else r.eval "plot(dx,dy, type=\"l\", col=\"#{@thread_colours[t.first]}\",ylim=c(0,#{density_max_y}), main='#{file}',xlab='position', ylab='density')" plot_open = true end end r.eval "dev.off()" if opts[:draw_legend] r.eval "png('#{opts[:draw_legend]}', width=#{opts[:width]}, height=#{opts[:height]})" colours = @thread_colours.each.sort.collect {|x| x.last}.join("','") names = @thread_colours.each.sort.collect {|x| x.first}.join("','") r.eval "plot(1,xlab="",ylab="",axes=FALSE)" r.eval "legend('top', c('#{names}'), lty=c(1),lwd=c(1),col=c('#{colours}'), ncol=4)" r.eval "dev.off()" end r.quit end public #Returns the instance variable @densities array of arrays [window, [density curve x values], [density curve y values] ]. The R function +density()+ is used to calculate the values. If @densities is nil when called this method will run the Bio::Util::Gngm#calculate_densities method and set @densities #With this method you cannot recalculate the densities after they have been done once. # #Options and defaults #- adjust = 1, -the kernel adjustment parameter for the R +density+ function def densities(adjust=1) @densities ||= calculate_densities(adjust) end public #Sets and returns the array of arrays [window, [density curve x values], [density curve y values] ] Calculates the density curve using the R function +density()+ Always sets @densities regardless of whether it contains anything or not so is useful for trying out adjustment values. #Ignores threads with fewer than 2 polymorphisms since density can't be computed with so few polymorphisms. # #Options and defaults #- adjust = 1, -the kernel adjustment parameter for the R +density+ function def calculate_densities(adjust=1) r = new_r densities = [] self.threads.each do |t| next if t.last.length < 2 ##length of density array is smaller or == threads, since too small windows are ignored... r.curr_win = t.last r.eval "d = density(curr_win,n=240,kernel=\"gaussian\", from=#{@snp_positions.first[0]}, to=#{@snp_positions.last[0]}, adjust=#{adjust})" densities << [t.first, r.pull("d$x"), r.pull("d$y")] end r.quit @densities = densities calculate_density_max_y ##need to re-do every time we get new densities densities end public #Draws the clustered bands that correspond to the expected and control window in a single PNG file +file+ # #Options and defaults #- :add_lines => nil -if an array of positions is provided eg +[100,345] , vertical lines will be drawn at these positions. Useful for indicating feature positions on the plot #- :width => 1000 -width of the PNG in pixels #- :height => 500 -height of the PNG in pixels def draw_bands(file="myfile.png", optsa={}) opts = @opts[:graphics].merge(optsa) pp optsa raise RuntimeError, "Can't draw threads until clustering is done" unless @clusters #uses R's standard plot functions. ##same as draw_threads, but skips threads that aren't on the bands lists ## r = new_r r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" plot_open = false self.densities.each do |t| if @control_band.include?(t[0]) or @expected_band.include?(t[0]) r.dx = t[1] r.dy = t[2] r.curr_win = t.last #r.eval "d = density(curr_win,n=240,kernel=\"gaussian\", from=#{@snp_positions.first[0]}, to=#{@snp_positions.last[0]})" if plot_open r.eval "lines(dx, dy, col=\"#{@thread_colours[t.first]}\")" else r.eval "plot(dx, dy, type=\"l\", col=\"#{@thread_colours[t.first]}\",ylim=c(0,#{density_max_y}), main='#{file}',xlab='position', ylab='density')" plot_open = true end end end label1 = "Control band: " + @control_band.min.to_s + " < ChD < " + @control_band.max.to_s label2 = "Expected band: " + @expected_band.min.to_s + " < ChD < " + @expected_band.max.to_s r.eval "legend('top', c('#{label1}','#{label2}'), lty=c(1,1),lwd=c(2.5,2.5),col=c('#{@thread_colours[@control_band.first]}','#{@thread_colours[@expected_band.first]}'))" if opts[:add_lines] and opts[:add_lines].instance_of?(Array) opts[:add_lines].each do |pos| r.eval "abline(v=#{pos})" end end r.eval "dev.off()" r.quit end public #Returns the array instance variable @clusters. The R function +kmeans()+ is used to calculate the clusters based on a correlation matrix of the density curves. If @clusters is nil when called this method will run the Bio::Util::Gngm#calculate_clusters method and set @clusters #With this method you cannot recalculate the clusters after they have been done once. # #Options and defaults #- :k => 9, -the number of clusters for the R +kmeans+ function #- :seed => false -set this to a number to make the randomized clustering reproducible #- :control_chd => 0.5 -the value of the control thread/window #- :expected_chd => 1.0 -the value of the expected thread/window #- :adjust => 1.0 -the kernel adjustment parameter for the R +density+ function def clusters(opts={}) @clusters ||= calculate_clusters(opts={}) end public #Calculates the k-means clusters of density curves (groups threads into bands), [density curve y values] ] Calculates the clusters using the R function +kmeans()+ Recalculates @densities as it does with Bio::Util::Gngm#calculate_densities, so clustering can be done without having to explicitly call Bio::Util::Gngm#calculate_densities. #Clusters are recalulated every time regardless of whether its been done before contains anything or not so is useful for trying out different values for the parameters. When clusters are calculated the expected and control #bands are compared with the Bio::Util::Gngm#calculate_signal method and the @signal array populated. Resets the instance variables @control_band, @expected_band, @signal, @peak_indices, @peak_y_values and @clusters # #Options and defaults #- :k => 9, -the number of clusters for the R +kmeans+ function #- :seed => false -set this to a number to make the randomized clustering reproducible #- :control_chd => 0.5 -the value of the control thread/window #- :expected_chd => 1.0 -the value of the expected thread/window #- :adjust => 1.0 -the kernel adjustment parameter for the R +density+ function #- :pseudo => false - force the densities into a single thread cluster when the number of distinct threads with SNPs is < the value of k. This is only useful in a situation where the spread of the statistic is very limited. EG for using mapped/unmapped mate pairs then almost all windows will have proportion 1.0 but a tiny number will be close to 0.5 with few other values considered. def calculate_clusters( opts={} ) options = {:k => 9, :seed => false, :adjust => 1, :control_chd => 0.5, :expected_chd => 1.0, :pseudo => false} options = options.merge(opts) if options[:pseudo] put_threads_into_individual_clusters(options) return end r = new_r names = [] name = "a" @control_band = nil #needs resetting as we are working with new clusters @expected_band = nil #needs resetting as we are working with new clusters @signal = nil #needs resetting as we are working with new clusters @peak_indices = nil #needs resetting as we are working with new cluster @peak_y_values = nil #needs resetting as we are working with new cluster self.calculate_densities(options[:adjust]).each do |d| density_array = d.last r.assign name, density_array ##although windows go in in numeric order, r wont allow numbers as names in data frames so we need a proxy names << "#{name}=#{name}" name = name.next end data_frame_command = "data = data.frame(" + names.join(",") + ")" r.eval data_frame_command r.eval "set.seed(#{options[:seed]})" if options[:seed] r.eval "k = kmeans(cor(data),#{options[:k]},nstart=1000)" @clusters = r.pull "k$cluster" ##clusters are returned in the order in densities r.quit ##now set the cluster colours.. colours = %w[#A6CEE3 #1F78B4 #B2DF8A #33A02C #FB9A99 #E31A1C #FDBF6F #FF7F00 #CAB2D6 #6A3D9A #FFFF99 #B15928] ci = 0 col_nums = {} ##hash of cluster numbers and colours @clusters.each_index do |i| if not col_nums[@clusters[i]] col_nums[@clusters[i]] = colours[ci] ci += 1 ci = 0 if ci > 11 end @thread_colours[self.densities[i].first] = col_nums[@clusters[i]] end @control_band = get_band(options[:control_chd]) @expected_band = get_band(options[:expected_chd]) calculate_signal end private ##gives each window/thread a seperate and arbitrary cluster, used when you suspect the statistic will not spread across all possible windows very well. Wont specifiy @control_band or @expected_band and therefore wont directly calculate the signal def put_threads_into_individual_clusters(options) @control_band = nil #needs resetting as we are working with new clusters @expected_band = nil #needs resetting as we are working with new clusters @signal = nil #needs resetting as we are working with new clusters @peak_indices = nil #needs resetting as we are working with new cluster @peak_y_values = nil #needs resetting as we are working with new cluster self.calculate_densities(options[:adjust]) @clusters = Array.new(@densities.length) {|x| 1 + x} ##now set the cluster colours.. colours = %w[#A6CEE3 #1F78B4 #B2DF8A #33A02C #FB9A99 #E31A1C #FDBF6F #FF7F00 #CAB2D6 #6A3D9A #FFFF99 #B15928] ci = 0 col_nums = {} ##hash of cluster numbers and colours @clusters.each_index do |i| if not col_nums[@clusters[i]] col_nums[@clusters[i]] = colours[ci] ci += 1 ci = 0 if ci > 11 end @thread_colours[self.densities[i].first] = col_nums[@clusters[i]] end #@control_band = get_band(options[:control_chd]) #@expected_band = get_band(options[:expected_chd]) #calculate_signal end ##returns an array of the names of the window threads in the control (heterophasic) band #def control_band(control=0.5) # puts "in control band with control = #{control}" # @control_band ||= get_band(control) #end ##returns an array of the names of the window threads in the expected (homophasic) band #def expected_band(expected=1.0) # @expected_band ||= get_band(expected) #end ##gets an array of windows that cluster with a given window public def get_band(window=1.0) ##because of the weird step rounding error we need to find the internal name of the window.. so find it from the list from the name the user ##expects it to be, may give more than one passing window so keep only first one.. windows = find_window(window) raise RuntimeError, "Couldnt find window #{window}, or window has no data to calculate: \n windows are #{self.densities.collect {|d| d.first} }" if windows.empty? ##if we have a window that is close enough to the specified window idx = find_index(windows.first) #find out which cluster the window is in cluster = self.clusters[idx] ##get the other windows in the same cluster, ie the band... band = [] self.clusters.each_index do |i| if self.clusters[i] == cluster band << self.densities[i].first end end band end public #Draws the contents of the @signal instance variable in a single PNG file +file+ def draw_signal(file="myfile.png", opts=@opts[:graphics]) #data.frame(bubs=data$bubbles_found,conf=data$bubbles_confirmed) r = new_r x_vals = self.densities[0][1] r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" r.x_vals = x_vals r.signal = self.signal r.eval "plot(x_vals,signal, type=\"l\", xlab='position', ylab='ratio of signals (expected / control ~ homo / hetero)', main='#{file}' )" r.eval "dev.off()" end private def print_signal end public #Returns the positions of the peaks in the signal curve calculated by Bio::Util::Gngm#get_peaks as an array def peaks @peak_indices.collect {|x| self.densities[0][1][x].to_f.floor} end public #Draws the peaks calculated from the signal curve by the R function +Peaks+ in Bio::Util::Gngm#calculate_peaks. Adds boxes of width +:range+ to each peak and annotates the limits. Options are set in the global options hash +:peaks+. #and relate to the Peaks function in R def draw_peaks(file="myfile.png",opts=@opts[:graphics]) opts_a = @opts[:peaks] opts_a.merge!(opts) opts = opts_a ##sigh ... #opts[:background] = opts[:background].to_s.upcase #opts[:markov] = opts[:markov].to_s.upcase self.get_peaks(opts) r = new_r #r.eval "suppressMessages ( library('Peaks') )" r.signal = self.signal r.x_vals = self.densities[0][1] r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" #r.eval "spec = SpectrumSearch(signal,#{opts[:sigma]},threshold=#{opts[:threshold]},background=#{opts[:background]},iterations=#{opts[:iterations]},markov=#{opts[:markov]},window=#{opts[:window]})" #peak_positions = r.pull "spec$pos" #y = r.pull "spec$y" r.y = @peak_y_values r.pos = @peak_indices r.eval "plot(x_vals,y, type=\"l\", xlab='position', ylab='Peaks', main='#{file}' )" @peak_indices.each do |peak| r.eval "rect(x_vals[#{peak}]-(#{opts[:range]/2}), 0, x_vals[#{peak}]+#{opts[:range]/2}, max(y), col=rgb(r=0,g=1,b=0, alpha=0.3) )" r.eval "text(x_vals[#{peak}]-(#{opts[:range]/2}),max(y) + 0.05, floor(x_vals[#{peak}]-(#{opts[:range]/2})) )" r.eval "text(x_vals[#{peak}]+(#{opts[:range]/2}), max(y) + 0.05, floor(x_vals[#{peak}]+(#{opts[:range]/2})) )" end r.eval "dev.off()" r.quit end #private #Calculates the position of peaks in the signal curve def get_peaks(opts=@opts[:peaks]) opts[:background] = opts[:background].to_s.upcase opts[:markov] = opts[:markov].to_s.upcase r = new_r r.eval "suppressMessages ( library('Peaks') )" r.signal = self.signal r.x_vals = self.densities[0][1] r.eval "spec = SpectrumSearch(signal,#{opts[:sigma]},threshold=#{opts[:threshold]},background=#{opts[:background]},iterations=#{opts[:iterations]},markov=#{opts[:markov]},window=#{opts[:window]})" @peak_indices = r.pull "spec$pos" if @peak_indices.instance_of?(Fixnum) @peak_indices = [@peak_indices] end @peak_y_values = r.pull "spec$y" r.quit end public #Returns an array of polymorphisms in each thread/window [[window, polymorphism count] ]. Useful for sparse polymorphism counts or over small regions where small polymorphism counts can cause artificially large peaks in density curves. def hit_count arr = [] self.threads.each do |thread| arr << [thread.first, thread.last.length] end arr end public #Draws a barplot of the number of polymorphisms in each thread/window in a single PNG file +file+ def draw_hit_count(file="myfile.png",opts=@opts[:graphics]) r = new_r wins = [] hits = [] self.threads.each do |thread| wins << thread.first if thread.last.empty? hits << 0.01 ##pseudovalue gets around the case where a thread has no hits... which messes up barplot in R else hits << thread.last.length end end r.wins = wins r.hits = hits r.eval "png('#{file}', width=#{opts[:width]}, height=#{opts[:height]})" r.eval "barplot(hits, names.arg=c(wins), xlab='window', log='y', ylab='number of hits', main='Number of Polymorphisms #{file}', col=rgb(r=0,g=1,b=1, alpha=0.3), na.rm = TRUE)" r.eval "dev.off()" end public #Returns an array of values representing the ratio of average of the expected threads/windows to the control threads/windows. Sets @signal, the signal curve. def calculate_signal r = new_r name = "a" control_names = [] expected_names = [] self.densities.each do |d| if @control_band.include?(d.first) density_array = d.last r.assign name, density_array ##although windows go in in numeric order, r wont allow numbers as names in data frames so we need a proxy control_names << "#{name}=#{name}" elsif @expected_band.include?(d.first) density_array = d.last r.assign name, density_array expected_names << "#{name}=#{name}" end name = name.next end data_frame_command = "control = data.frame(" + control_names.join(",") + ")" r.eval data_frame_command r.eval "control_mean = apply(control, 1, function(ecks) mean((as.numeric(ecks))) )" data_frame_command = "expected = data.frame(" + expected_names.join(",") + ")" r.eval data_frame_command r.eval "expected_mean = apply(expected, 1, function(ecks) mean((as.numeric(ecks))) )" r.eval "signal = expected_mean / control_mean" signal = r.pull "signal" r.quit @signal = signal end public def signal @signal ||= calculate_signal end ##finds the index of a window in the densties array private def find_index(window) self.densities.index {|x| x.first == window} end private #finds the windows internal name, taking into account the Ruby rounding error def find_window(number) self.densities.collect {|d| d.first if d.first == number or (d.first >= number - ERROR_MARGIN and d.first <= number + ERROR_MARGIN) }.compact end private #Returns a new rinruby object def new_r r = RinRuby.new(echo = false, interactive = false) r.eval "options(warn=-1)" return r end private #returns an array of arrays of known variants #file: #chr1 500 A G #chr2 1000 ATGTTA #chr3 1500 . TTGGA # returns [["chr1", "500", "A", "G"], ["chr2", "1000", "ATG", "TTA"], ["chr3", "1500", ".", "TTGGA"]] def parse_known_variants(file) File.open(file, "r").readlines.collect {|x| x.chomp.split("\t")} end public #Deletes everything from self.snp_positions not mentioned by position in self.known_variants. Directly modifies self.snp_positions def keep_known_variants(file=nil) raise "file of known variants not provided and @known_variants is nil" if @known_variants.nil? and file.nil? @known_variants = parse_known_variants(file) if @known_variants.nil? and file @snp_positions.each do |snp| end end end end end