diff cravat_convert/vcf_converter.py @ 22:0ca746ab1151 draft

Uploaded
author in_silico
date Wed, 18 Jul 2018 10:18:06 -0400
parents c042835a7163
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/cravat_convert/vcf_converter.py	Wed Jul 18 10:18:06 2018 -0400
@@ -0,0 +1,243 @@
+"""
+A module originally obtained from the cravat package. Modified to use in the vcf
+converter galaxy tool.
+
+
+Register of changes made (Chris Jacoby):
+    1) Changed imports as galaxy tool won't have access to complete cravat python package
+    2) Defined BadFormatError in BaseConverted file, as I didn't have the BadFormatError module
+"""
+
+from base_converter import BaseConverter, BadFormatError
+import re
+
+class CravatConverter(BaseConverter):
+    
+    def __init__(self):
+        self.format_name = 'vcf'
+        self.samples = []
+        self.var_counter = 0
+        self.addl_cols = [{'name':'phred',
+                           'title':'Phred',
+                           'type':'string'},
+                          {'name':'filter',
+                           'title':'VCF filter',
+                           'type':'string'},
+                          {'name':'zygosity',
+                           'title':'Zygosity',
+                           'type':'string'},
+                          {'name':'alt_reads',
+                           'title':'Alternate reads',
+                           'type':'int'},
+                          {'name':'tot_reads',
+                           'title':'Total reads',
+                           'type':'int'},
+                          {'name':'af',
+                           'title':'Variant allele frequency',
+                           'type':'float'}]
+    
+    def check_format(self, f): 
+        return f.readline().startswith('##fileformat=VCF')
+    
+    def setup(self, f):
+        
+        vcf_line_no = 0
+        for line in f:
+            vcf_line_no += 1
+            if len(line) < 6:
+                continue
+            if line[:6] == '#CHROM':
+                toks = re.split('\s+', line.rstrip())
+                if len(toks) > 8:
+                    self.samples = toks[9:]
+                break
+    
+    def convert_line(self, l):
+        if l.startswith('#'): return None
+        self.var_counter += 1
+        toks = l.strip('\r\n').split('\t')
+        all_wdicts = []
+        if len(toks) < 8:
+            raise BadFormatError('Wrong VCF format')
+        [chrom, pos, tag, ref, alts, qual, filter, info] = toks[:8]
+        if tag == '':
+            raise BadFormatError('ID column is blank')
+        elif tag == '.':
+            tag = 'VAR' + str(self.var_counter)
+        if chrom[:3] != 'chr':
+            chrom = 'chr' + chrom
+        alts = alts.split(',')
+        len_alts = len(alts)
+        if len(toks) == 8:
+            for altno in range(len_alts):
+                wdict = None
+                alt = alts[altno]
+                newpos, newref, newalt = self.extract_vcf_variant('+', pos, ref, alt)
+                wdict = {'tags':tag,
+                         'chrom':chrom,
+                         'pos':newpos,
+                         'ref_base':newref,
+                         'alt_base':newalt,
+                         'sample_id':'no_sample',
+                         'phred': qual,
+                         'filter': filter}
+                all_wdicts.append(wdict)
+        elif len(toks) > 8:
+            sample_datas = toks[9:]
+            genotype_fields = {}
+            genotype_field_no = 0
+            for genotype_field in toks[8].split(':'):
+                genotype_fields[genotype_field] = genotype_field_no
+                genotype_field_no += 1
+            if not ('GT' in genotype_fields):
+                raise BadFormatError('No GT Field')
+            gt_field_no = genotype_fields['GT']
+            for sample_no in range(len(sample_datas)):
+                sample = self.samples[sample_no]
+                sample_data = sample_datas[sample_no].split(':')
+                gts = {}
+                for gt in sample_data[gt_field_no].replace('/', '|').split('|'):
+                    if gt == '.':
+                        continue
+                    else:
+                        gts[int(gt)] = True
+                for gt in sorted(gts.keys()):
+                    wdict = None
+                    if gt == 0:
+                        continue
+                    else:
+                        alt = alts[gt - 1]
+                        newpos, newref, newalt = self.extract_vcf_variant('+', pos, ref, alt)
+                        zyg = self.homo_hetro(sample_data[gt_field_no])
+                        depth, alt_reads, af = self.extract_read_info(sample_data, gt, gts, genotype_fields)
+                            
+                        wdict = {'tags':tag,
+                                 'chrom':chrom,
+                                 'pos':newpos,
+                                 'ref_base':newref,
+                                 'alt_base':newalt,
+                                 'sample_id':sample,
+                                 'phred': qual,
+                                 'filter': filter,
+                                 'zygosity': zyg,
+                                 'tot_reads': depth,
+                                 'alt_reads': alt_reads,
+                                 'af': af,                                
+                                 }
+                        all_wdicts.append(wdict)
+        return all_wdicts
+ 
+    #The vcf genotype string has a call for each allele separated by '\' or '/'
+    #If the call is the same for all allels, return 'hom' otherwise 'het'
+    def homo_hetro(self, gt_str):
+        if '.' in gt_str:
+            return '';
+        
+        gts = gt_str.strip().replace('/', '|').split('|')
+        for gt in gts:
+            if gt != gts[0]:
+                return 'het'
+        return 'hom'            
+                        
+    #Extract read depth, allele count, and allele frequency from optional VCR information
+    def extract_read_info (self, sample_data, gt, gts, genotype_fields): 
+        depth = ''
+        alt_reads = ''
+        ref_reads = ''
+        af = ''
+        
+        #AD contains 2 values usually ref count and alt count unless there are 
+        #multiple alts then it will have alt 1 then alt 2.
+        if 'AD' in genotype_fields and genotype_fields['AD'] <= len(sample_data): 
+            if 0 in gts.keys():
+                #if part of the genotype is reference, then AD will have #ref reads, #alt reads
+                ref_reads = sample_data[genotype_fields['AD']].split(',')[0]
+                alt_reads = sample_data[genotype_fields['AD']].split(',')[1]
+            elif gt == max(gts.keys()):    
+                #if geontype has multiple alt bases, then AD will have #alt1 reads, #alt2 reads
+                alt_reads = sample_data[genotype_fields['AD']].split(',')[1]
+            else:
+                alt_reads = sample_data[genotype_fields['AD']].split(',')[0]                            
+                             
+        if 'DP' in genotype_fields and genotype_fields['DP'] <= len(sample_data): 
+            depth = sample_data[genotype_fields['DP']] 
+        elif alt_reads != '' and ref_reads != '':
+            #if DP is not present but we have alt and ref reads count, dp = ref+alt
+            depth = int(alt_reads) + int(ref_reads)   
+
+        if 'AF' in genotype_fields and genotype_fields['AF'] <= len(sample_data):
+            af = float(sample_data[genotype_fields['AF']] )
+        elif depth != '' and alt_reads != '':
+            #if AF not specified, calc it from alt and ref reads
+            af = float(alt_reads) / float(depth)
+ 
+        return depth, alt_reads, af
+            
+    def extract_vcf_variant (self, strand, pos, ref, alt):
+
+        reflen = len(ref)
+        altlen = len(alt)
+        
+        # Returns without change if same single nucleotide for ref and alt. 
+        if reflen == 1 and altlen == 1 and ref == alt:
+            return pos, ref, alt
+        
+        # Trimming from the start and then the end of the sequence 
+        # where the sequences overlap with the same nucleotides
+        new_ref2, new_alt2, new_pos = \
+            self.trimming_vcf_input(ref, alt, pos, strand)
+                
+        if new_ref2 == '':
+            new_ref2 = '-'
+        if new_alt2 == '':
+            new_alt2 = '-'
+        
+        return new_pos, new_ref2, new_alt2
+    
+    # This function looks at the ref and alt sequences and removes 
+    # where the overlapping sequences contain the same nucleotide.
+    # This trims from the end first but does not remove the first nucleotide 
+    # because based on the format of VCF input the 
+    # first nucleotide of the ref and alt sequence occur 
+    # at the position specified.
+    #     End removed first, not the first nucleotide
+    #     Front removed and position changed
+    def trimming_vcf_input(self, ref, alt, pos, strand):
+        pos = int(pos)
+        reflen = len(ref)
+        altlen = len(alt)
+        minlen = min(reflen, altlen)
+        new_ref = ref
+        new_alt = alt
+        new_pos = pos
+        # Trims from the end. Except don't remove the first nucleotide. 
+        # 1:6530968 CTCA -> GTCTCA becomes C -> GTC.
+        for nt_pos in range(0, minlen - 1): 
+            if ref[reflen - nt_pos - 1] == alt[altlen - nt_pos - 1]:
+                new_ref = ref[:reflen - nt_pos - 1]
+                new_alt = alt[:altlen - nt_pos - 1]
+            else:
+                break    
+        new_ref_len = len(new_ref)
+        new_alt_len = len(new_alt)
+        minlen = min(new_ref_len, new_alt_len)
+        new_ref2 = new_ref
+        new_alt2 = new_alt
+        # Trims from the start. 1:6530968 G -> GT becomes 1:6530969 - -> T.
+        for nt_pos in range(0, minlen):
+            if new_ref[nt_pos] == new_alt[nt_pos]:
+                if strand == '+':
+                    new_pos += 1
+                elif strand == '-':
+                    new_pos -= 1
+                new_ref2 = new_ref[nt_pos + 1:]
+                new_alt2 = new_alt[nt_pos + 1:]
+            else:
+                new_ref2 = new_ref[nt_pos:]
+                new_alt2 = new_alt[nt_pos:]
+                break  
+        return new_ref2, new_alt2, new_pos
+
+
+if __name__ == "__main__":
+    c = CravatConverter()
\ No newline at end of file