Пример #1
0
def main():
    """ getting the git hash in python: http://stackoverflow.com/a/14989911/1735942"""
    label = subprocess.check_output(["git", "rev-parse", "HEAD"])
    today=datetime.datetime.today()
    datestr=today.strftime("20%y%m%d")
    usage = "usage: %prog [options] my.bam "
    parser = OptionParser(usage)
    parser.add_option("--bed", type="string", dest="bedfile", help="bed file with coordinates")
    parser.add_option("--tbf", type="string", dest="tbfile", help=" *.2bit file of the reference sequence")
    parser.add_option("--ped", type="string", dest="pedfile", help= " pedfile of the samples you are analyzing")
    parser.add_option("--min-nonref-count", dest="minAlt", default=2, help="minimum observation of nonref allele for genotype inference (default 2)")
    (options, args)=parser.parse_args()
    
    commandline=";".join(sys.argv)
    
    bamfile=args[0]
    bamfilebasename=return_file_basename(bamfile)
    vcfoutput=".".join(['pgmsnp',bamfilebasename, datestr, 'vcf'])
    vcfh=open(vcfoutput,'w')
    ALPHABET=['AA', 'AC', 'AG', 'AT', 'CC', 'CG', 'CT', 'GG', 'GT', 'TT']
   
    twobit=bx.seq.twobit.TwoBitFile( open( options.tbfile  ) )
    bedobj=Bedfile(options.bedfile)
    pybamfile=pysam.Samfile( bamfile, "rb" )
    pedfile=Pedfile(options.pedfile)
    pedfile.parsePedfile()

    samplenames=pedfile.returnIndivids()
    samplestring="\t".join(samplenames)
    

    #vcf metainfolines
    vcf_metalines=[]
    vcf_metalines.append ( "##fileformat=VCFv4.1")
    vcf_metalines.append( "##fileDate="+datestr )
    vcf_metalines.append("##testGeneticFactor="+commandline)
    vcf_metalines.append("##version="+label.strip())
    vcf_metalines.append("##reference="+options.tbfile)
    vcf_metalines.append("##pedfile="+options.pedfile)
    vcf_metalines.append("##bedfile="+options.bedfile)
    vcf_metalines.append("##bamfile="+bamfile)
    vcf_metalines.append("##INFO=<ID=NS,Number=1,Type=Integer,Description=\"Number of Samples With Data\">")
    vcf_metalines.append( "##INFO=<ID=DP,Number=1,Type=Integer,Description=\"Total Depth\">" )
    vcf_metalines.append( "##FORMAT=<ID=GT,Number=1,Type=String,Description=\"Genotype\">" )
    vcf_metalines.append( "##FORMAT=<ID=GQ,Number=1,Type=Integer,Description=\"Genotype Quality\">" )
    vcf_metalines.append( "##FORMAT=<ID=DP,Number=1,Type=Integer,Description=\"Read Depth\">" )
    vcf_column_headers=["#CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO", "FORMAT",samplestring]
    vcf_metalines.append( "\t".join(vcf_column_headers))

    vcf_metaline_output="\n".join(vcf_metalines)
    vcfh.write(vcf_metaline_output+"\n")
    #print vcf_metaline_output



    #print samplenames

    #return the complete enumeration of all possible offspring genotype priors
    punnetValues=returnPunnetValues(4)

    # Pfactor gives us a pileup iterator
    Pfactory=PileupFactory(pybamfile,bedobj)



    ##CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO", "FORMAT",samplestring

    for pileup_data_obj in Pfactory.yieldPileupData():
        
        refDepth=defaultdict(lambda: defaultdict(int))
        altDepth=defaultdict(lambda: defaultdict(int))
        pileup_column_pos=pileup_data_obj.getPileupColumnPos()
        pileup_column_chrom=pileup_data_obj.getChrom()
        refsequence=twobit[pileup_column_chrom][pileup_column_pos:pileup_column_pos+1] #this is the reference base
        refDepth=defaultdict(lambda: defaultdict(int))
        altDepth=defaultdict(lambda: defaultdict(int))
        skipsite=False
        #print 'Pileup position: ', pileup_data_obj, "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence])
        #print
        
        qual="."
        filter='.'
        siteDP="DP="+str(pileup_data_obj.getPileupDepth())
        altDP=0
        sampleDepth=defaultdict(int)


        #lets make our genetic network here:
        pgmNetwork=PgmNetworkFactory(options.pedfile,pileup_column_chrom,pileup_column_pos, refsequence,punnetValues)
        totalSize=pgmNetwork.returnTotalSize()
        #print "totalSize: ", totalSize
        observedSamples=[]
        sampleNames=set(pgmNetwork.getSampleNames())
        #print "pgmNetwork factor list: "
        #pgmNetwork.printFactorList()
        pgmFactorList=pgmNetwork.getFactorList()
        #print "++++"
        
        for (sample, pileup_sample_data) in pileup_data_obj.yieldSamplePileupData():
            sampleDepth[sample]=len(pileup_sample_data)
            observedSamples.append(sample)
            sample_idx=pgmNetwork.getSampleNamePedIndex(sample)
            
            for data in pileup_sample_data:
                if data.basecall != refsequence: altDepth[data.sample][data.basecall]+=1
                if data.basecall == refsequence: refDepth[data.sample][data.basecall]+=1
            
       
            
            value=calculateGLL(pileup_sample_data)
            

            pgmFactorList[sample_idx + totalSize].setVal(value)
            #we initially get the log-likelihood, but go back to probablity space first
            pgmFactorList[sample_idx + totalSize] = ExpFactor( pgmFactorList[sample_idx + totalSize] )
            #print pgmFactorList[sample_idx + totalSize]
            
            #GLFactor = Factor( [readVar, genoVar], [1,10], [], 'read_phenotype | genotype ')
            #gPrior=LogFactor( returnGenotypePriorFounderFactor(sequence,['A','C','G','T'], genoVar) )
            #print
        
    
        if sum(chain.from_iterable(d.itervalues() for d in altDepth.itervalues())) < options.minAlt: 
       #print pileup_data_obj.getPileupColumnPos(), altDP
            continue
        
        observedSamples=set(observedSamples)
        unobservedSamples=sampleNames-observedSamples
        unobservedIdxs=[ pgmNetwork.getSampleNamePedIndex(sample) for sample in unobservedSamples  ]
        #for samples lacking read data, delete them from the list of
        for idx in unobservedIdxs:
            #del pgmFactorList[idx + totalSize]
            value=calculateNoObservationsGL()
            #pdb.set_trace()
            pgmFactorList[idx + totalSize].setVal(value)

        #print pgmFactorList
        #for f in pgmFactorList:
        #    print f
        #continue
        #print "====="
        siteNS="NS="+str(len(observedSamples))
        infoString=";".join([siteNS, siteDP])
        
        
        
        #for f in pgmFactorList:
        #    print f
        #    print

        #pdb.set_trace()
        #cTree=CreatePrunedInitCtree(pgmFactorList)
        #print 'cTree constructed, pruned, and initalized potential'


        #G=nx.from_numpy_matrix( cTree.getEdges() )
        #nx.draw_shell(G)
        #plt.show()


        #print cTree

        #get the max marginal factors
        MAXMARGINALS=ComputeExactMarginalsBP(pgmFactorList, [], 1)
        
        #MARGINALS= ComputeExactMarginalsBP( pgmFactorList)
        #this log normalizes the data
        for m in MAXMARGINALS:
            #print m.getVal()
            m.setVal( np.log( lognormalize(m.getVal()   )   ) )
            #print np.sum( lognormalize(m.getVal() ) )
            #print'lognormalize: ',  lognormalize(m.getVal() )
            #print m.getVal()
            #print
        #    print  m.getVal()
        #    print
        #print "==="
        #get the max value of each factor
        
        
                
                           
        MAXvalues=[    max(m.getVal().tolist() )  for m in MAXMARGINALS  ]
        
        
        #MAXvalues_phred = [ PhredScore(x) for x in MAXvalues ]
        """ phred scale the error prob. so we take the 1-posterior, and phred scale it
            Note we have expoentiate the values then take the complement, then phred scale """
        phred_gq= [ round(PhredScore(x),2) for x in (1-np.exp(MAXvalues[0:totalSize])) ]
        #MAXvalues_str= [str(x) for x in MAXvalues]
        phred_str=[str(x) for x in phred_gq]
        #print "maxvalues: ",  MAXvalues[0:totalSize]
        #this is the decoding, where we get teh assignment of the variable
        # that has the max marginoal value
        MAPAssignment=MaxDecoding( MAXMARGINALS  )
        #print "MAPAssignment: ",  MAPAssignment[0:3]
        #print totalSize
        #for idx in range(totalSize):
        #    print ALPHABET[ MAPAssignment[idx] ]
        #print MAPAssignment

        #we convert from  variable assignment in the factor to genotype space
        sampleNames=pgmNetwork.returnGenotypeFactorIdxSampleNames()
      
        sampleDepths=[]
        for sample in sampleNames:
            sampleDepths.append(str(sampleDepth[sample]))
        MAPgenotypes=[ALPHABET[ i ] for i in  MAPAssignment[0:totalSize] ]
        #print MAPgenotypes

        alt=determineAltBases( MAPgenotypes, [refsequence] )
        numericalMAPgenotypes=[ numericalGenotypes(refsequence,alt,map_g) for map_g in MAPgenotypes ]
        #print numericalMAPgenotypes
        site_info="\t".join([pileup_column_chrom, str(pileup_column_pos+1), ".", refsequence,alt,qual,filter,infoString, "GT:GQ:DP"])
        #zip the genotype assignment and its log-probability together
        #genotypedata=zip(MAPgenotypes,sampleDepths,MAXvalues[0:totalSize])
        #genotypedata=zip(MAPgenotypes,sampleDepths)
        #genotypedata=zip(numericalMAPgenotypes,sampleDepths,  MAXvalues_str[0:totalSize] )
        genotypedata=zip(numericalMAPgenotypes,phred_str, sampleDepths, )
        #genotypedata=zip(numericalMAPgenotypes,sampleDepths)
        
        genotypedata=[ ":".join(list(i)) for i in  genotypedata ]
        #print genotypedata

        #print  genotypedata
        #print sampleGenotypes
        #print "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence]) + "\t" +"\t".join(MAPgenotypes)
        if int(pileup_column_pos) % 10000 == 0:
            sys.stdout.write("processed 10 kbp ...\n")
        #print "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence]) + "\t" +"\t".join(numericalMAPgenotypes)
        #print site_info
        vcfgenotypeoutput="\t".join(genotypedata)
        output=site_info+ "\t"+vcfgenotypeoutput
        vcfh.write(output+"\n")
Пример #2
0
def main():
    """ getting the git hash in python: http://stackoverflow.com/a/14989911/1735942"""
    #label = subprocess.check_output(["git", "rev-parse", "HEAD"])
    today=datetime.datetime.today()
    datestr=today.strftime("20%y%m%d")
    usage = "usage: %prog [options] my.bam "
    parser = OptionParser(usage)
    parser.add_option("--bed", type="string", dest="bedfile", help="bed file with coordinates")
    parser.add_option("--tbf", type="string", dest="tbfile", help=" *.2bit file of the reference sequence")
    parser.add_option("--ped", type="string", dest="pedfile", help= " pedfile of the samples you are analyzing")
    parser.add_option("--min-nonref-count", type="int", dest="minAlt", default=2, help="minimum observation of nonref allele for genotype inference (default 2)")
    parser.add_option("--mapq", dest="mapq", type="int", default=30, help="exclude read if  mapping quality is less than Q default:30 ")
    parser.add_option("--bq", dest="bq", type="int", default=25, help="exclude base if basequality is less than Q: default 25")
    
    parser.add_option("--emitall", action="store_true", dest="emitall", help="emit all sites, even those without min alt threshold", default=False)
    parser.add_option("--debug", action="store_true", dest="debug", default=False)
    (options, args)=parser.parse_args()
    
    commandline=";".join(sys.argv)
    
    bamfile=args[0]
    #print options.bedfile
   
    bamfilebasename=return_file_basename(bamfile)
    bedfilebasename=return_file_basename(options.bedfile)
    vcfoutput=".".join(['pgmsnp',bamfilebasename,bedfilebasename, datestr, 'vcf'])
    
    
    if options.debug == True:
        prettybaseoutput=".".join(['pgmsnp',bamfilebasename, datestr,'pretty' ])
        try:
            os.remove(prettybaseoutput)
        except OSError:
            pass
        prettyfh=open(prettybaseoutput, 'a')    
    
    vcfh=open(vcfoutput,'w')
    DNA_ALPHABET= ['A', 'C', 'G' ,'T']
    ALPHABET=['AA', 'AC', 'AG', 'AT', 'CC', 'CG', 'CT', 'GG', 'GT', 'TT']
  
    twobit=bx.seq.twobit.TwoBitFile( open( options.tbfile  ) )
    bedobj=Bedfile(options.bedfile)
    pybamfile=pysam.Samfile( bamfile, "rb" )
    pedfile=Pedfile(options.pedfile)
    pedfile.parsePedfile()

    samplenames=pedfile.returnIndivids()
    
    samplestring="\t".join(samplenames)
    

    #vcf metainfolines
    vcf_metalines=[]
    vcf_metalines.append ( "##fileformat=VCFv4.1")
    vcf_metalines.append( "##fileDate="+datestr )
    vcf_metalines.append("##pgmsnp.py="+commandline)
    #vcf_metalines.append("##version="+label.strip())
    vcf_metalines.append("##reference="+options.tbfile)
    vcf_metalines.append("##pedfile="+options.pedfile)
    vcf_metalines.append("##bedfile="+options.bedfile)
    vcf_metalines.append("##bamfile="+bamfile)
    vcf_metalines.append("##INFO=<ID=NS,Number=1,Type=Integer,Description=\"Number of Samples With Data\">")
    vcf_metalines.append( "##INFO=<ID=DP,Number=1,Type=Integer,Description=\"Total Depth\">" )
    vcf_metalines.append( "##FORMAT=<ID=GT,Number=1,Type=String,Description=\"Genotype\">" )
    vcf_metalines.append( "##FORMAT=<ID=GQ,Number=1,Type=Integer,Description=\"Genotype Quality\">" )
    vcf_metalines.append( "##FORMAT=<ID=DP,Number=1,Type=Integer,Description=\"Read Depth\">" )
    #vcf_metalines.apppend("##FILTER=<ID=LowQual,Description=\"Low quality\">")
    vcf_column_headers=["#CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO", "FORMAT",samplestring]
    vcf_metalines.append( "\t".join(vcf_column_headers))

    vcf_metaline_output="\n".join(vcf_metalines)
    vcfh.write(vcf_metaline_output+"\n")
    #print vcf_metaline_output



    #print samplenames

    #return the complete enumeration of all possible offspring genotype priors
    punnetValues=returnPunnetValues(4)

    # Pfactor gives us a pileup iterator
    Pfactory=PileupFactory(pybamfile,bedobj)



    ##CHROM", "POS", "ID", "REF", "ALT", "QUAL", "FILTER", "INFO", "FORMAT",samplestring

    for pileup_data_obj in Pfactory.yieldPileupData(options.mapq, options.bq):
        
        
        refDepth=defaultdict(lambda: defaultdict(int))
        altDepth=defaultdict(lambda: defaultdict(int))
        pileup_column_pos=pileup_data_obj.getPileupColumnPos()
        pileup_column_chrom=pileup_data_obj.getChrom()
        refsequence=twobit[pileup_column_chrom][pileup_column_pos:pileup_column_pos+1] #this is the reference base
        refDepth=defaultdict(lambda: defaultdict(int))
        altDepth=defaultdict(lambda: defaultdict(int))
        skipsite=False
        #print 'Pileup position: ', pileup_data_obj, "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence])
        if refsequence not in DNA_ALPHABET: continue
        #print
        
        qual="."
        filter='.'
        siteDP="DP="+str(pileup_data_obj.getPileupDepth())
        altDP=0
        sampleDepth=defaultdict(int)


        #lets make our genetic network here:
        pgmNetwork=PgmNetworkFactory(options.pedfile,pileup_column_chrom,pileup_column_pos, refsequence,punnetValues)
        totalSize=pgmNetwork.returnTotalSize()
        #print "totalSize: ", totalSize
        observedSamples=[]
        sampleNames=set(pgmNetwork.getSampleNames())
        #print "pgmNetwork factor list: "
        #pgmNetwork.printFactorList()
        pgmFactorList=pgmNetwork.getFactorList()
        #print "++++"
        
        for (sample, pileup_sample_data) in pileup_data_obj.yieldSamplePileupData():
            sampleDepth[sample]=len(pileup_sample_data)
            observedSamples.append(sample)
            sample_idx=pgmNetwork.getSampleNamePedIndex(sample)
            
            for data in pileup_sample_data:
                if data.basecall != refsequence: altDepth[data.sample][data.basecall]+=1
                if data.basecall == refsequence: refDepth[data.sample][data.basecall]+=1
            
       
            
            value=calculateGLL(pileup_sample_data)
            #pdb.set_trace()
            

            pgmFactorList[sample_idx + totalSize].setVal(value)
            #pdb.set_trace()
            """we initially get the log-likelihood, but go back to probablity space first """
            #pgmFactorList[sample_idx + totalSize] = ExpFactor( pgmFactorList[sample_idx + totalSize] )
            pgmFactorList[sample_idx + totalSize] = ExpFactorNormalize( pgmFactorList[sample_idx + totalSize] )
            
            #print pgmFactorList[sample_idx + totalSize]
            
            #GLFactor = Factor( [readVar, genoVar], [1,10], [], 'read_phenotype | genotype ')
            #gPrior=LogFactor( returnGenotypePriorFounderFactor(sequence,['A','C','G','T'], genoVar) )
            #print
        
       
        """ in order for genotype inference to continue, minAlt value must be passed
            If it isn't and emitall is True, then we do genotype inference. Otherwise continue""" 
        if sum(chain.from_iterable(d.itervalues() for d in altDepth.itervalues())) < options.minAlt and options.emitall == False: 
       
            continue
        
        observedSamples=set(observedSamples)
        unobservedSamples=sampleNames-observedSamples
        unobservedIdxs=[ pgmNetwork.getSampleNamePedIndex(sample) for sample in unobservedSamples  ]
        #for samples lacking read data, delete them from the list of
        for idx in unobservedIdxs:
            #del pgmFactorList[idx + totalSize]
            value=calculateNoObservationsGL()
            #pdb.set_trace()
            pgmFactorList[idx + totalSize].setVal(value)

        #print pgmFactorList
        #for f in pgmFactorList:
        #    print f
        #continue
        #print "====="
        siteNS="NS="+str(len(observedSamples))
        infoString=";".join([siteNS, siteDP])
        
        
        
        #for f in pgmFactorList:
        #    print f
        #    print

        #pdb.set_trace()
        #cTree=CreatePrunedInitCtree(pgmFactorList)
        #print 'cTree constructed, pruned, and initalized potential'


        #G=nx.from_numpy_matrix( cTree.getEdges() )
        #nx.draw_shell(G)
        #plt.show()


        #print cTree

        #get the max marginal factors
        if options.debug == True:
            sys.stderr.write("computing exact marginals and joint from clique tree...\n")
        
        (MAXMARGINALS,JOINT)=ComputeExactMarginalsBP(pgmFactorList, [], 1, 1)
        
        sampleNames=pgmNetwork.returnGenotypeFactorIdxSampleNames()
        #pdb.set_trace()
        if options.debug == True:
            posterior_genotypes_values(MAXMARGINALS, ALPHABET,samplenames, "\t".join([pileup_column_chrom, str(pileup_column_pos+1)]), prettyfh)
        
        
        #this log normalizes the data
        for m in MAXMARGINALS:
            #print m
            #print m.getVal()
            #pdb.set_trace()
            m.setVal( np.log( m.getVal() / np.sum(m.getVal() )) )
            #m.setVal( np.log( lognormalize(m.getVal()   )   ) )
            #print np.sum( lognormalize(m.getVal() ) )
            #print'lognormalize: ',  lognormalize(m.getVal() )
            #print m.getVal()
            #print
        #    print  m.getVal()
        #    print
        #print "==="
        #get the max value of each factor
        
        #print len(MAXMARGINALS)
                
        """ max values will hold the max; assuming its unique. 
            get a list of the the values as well """  
        
        #for m in MAXMARGINALS:
        #    print m
        #    print "total size: ", totalSize
        #    """ ask yourself, is the  max unique? """
        #    print max(m.getVal().tolist())
        #    print
                          
        MAXvalues=[    max(m.getVal().tolist() )  for m in MAXMARGINALS  ]
        #pdb.set_trace()
        
        #MAXvalues_phred = [ PhredScore(x) for x in MAXvalues ]
        """ this code deals with calculating genotype quality GQ .... """
        """ phred scale the error prob. so we take the 1-posterior, and phred scale it
            Note we have expoentiate the values then take the complement, then phred scale """
        phred_gq= [ round(PhredScore(x),2) for x in (1-np.exp(MAXvalues[0:totalSize])) ]
        #MAXvalues_str= [str(x) for x in MAXvalues]
        phred_str=[str(x) for x in phred_gq]
        #print "maxvalues: ",  MAXvalues[0:totalSize]
        #this is the decoding, where we get teh assignment of the variable
        # that has the max marginoal value
        
        MAPAssignment=MaxDecoding( MAXMARGINALS  )
        #MAPAssignment=MaxDecodingNonUniq( MAXMARGINALS  )
        #pdb.set_trace()
        
        #print "MAPAssignment: ",  MAPAssignment[0:3]
        #print totalSize
        #for idx in range(totalSize):
        #    print ALPHABET[ MAPAssignment[idx] ]
        #print MAPAssignment

        #we convert from  variable assignment in the factor to genotype space
        sampleNames=pgmNetwork.returnGenotypeFactorIdxSampleNames()
      
        sampleDepths=[]
        for sample in sampleNames:
            sampleDepths.append(str(sampleDepth[sample]))
        MAPgenotypes=[ALPHABET[ i ] for i in  MAPAssignment[0:totalSize] ]
        #print MAPgenotypes

        alt=determineAltBases( MAPgenotypes, [refsequence] )
        
        dimension=np.prod(JOINT.getCard())
        strides=variableStride(JOINT)
        RR_genotype=refsequence+refsequence
        idx=IndexToAssignment ( np.arange(np.prod(JOINT.getCard()) ), JOINT.getCard() )-1
        g_idx= genotypeToIndex( RR_genotype, ploidy=2)
        """ conviently enough, the index of the all ref assignment is genotype idx repeated however many times there are samples """
        idx_string=str(g_idx)*totalSize
        
        idx_numbr=int(idx_string)
        allrefs_ass=generateAllReferenceGenotypesAssignment(g_idx, totalSize)
        
        if np.array_equal(idx[idx_numbr], allrefs_ass) == False:
            print "==="
            print "value assignment doesn't equal all reference assignment"
            print idx[idx_numbr]
            print allrefs_ass
            print "==="
            sys.exit(1)
        #idx_all_refs_ass=IndexOfAssignment(JOINT, strides, allref_ass )
        else:
            val=JOINT.getVal()[idx_numbr]
        
        #pdb.set_trace()
        
        """ Phred score takes in the error probability
        In this case the error prob is the prob that everyone is homozygous reference
        which is the val we get from the joint distribution factor above """
        QUAL=round(PhredScore(val),2)
       
        """ emit numerical genotypes as per vcf convention """
        numericalMAPgenotypes=[ numericalGenotypes(refsequence,alt,map_g) for map_g in MAPgenotypes ]
        #print numericalMAPgenotypes
        site_info="\t".join([pileup_column_chrom, str(pileup_column_pos+1), ".",refsequence,alt,str(QUAL),filter,infoString, "GT:GQ:DP"])
        #print site_info
        #pdb.set_trace()
        #site_info="\t".join([pileup_column_chrom, str(pileup_column_pos+1), ".", refsequence,alt,qual,filter,infoString, "GT:GQ:DP"])
        #zip the genotype assignment and its log-probability together
        #genotypedata=zip(MAPgenotypes,sampleDepths,MAXvalues[0:totalSize])
        #genotypedata=zip(MAPgenotypes,sampleDepths)
        #genotypedata=zip(numericalMAPgenotypes,sampleDepths,  MAXvalues_str[0:totalSize] )
        genotypedata=zip(numericalMAPgenotypes,phred_str, sampleDepths, )
        #genotypedata=zip(numericalMAPgenotypes,sampleDepths)
        
        genotypedata=[ ":".join(list(i)) for i in  genotypedata ]
        #print genotypedata

        #print  genotypedata
        #print sampleGenotypes
        #print "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence]) + "\t" +"\t".join(MAPgenotypes)
        if int(pileup_column_pos) % 10000 == 0:
            sys.stdout.write("processed 10 kbp ...\n")
        #print "\t".join( [pileup_column_chrom,  str(pileup_column_pos), refsequence]) + "\t" +"\t".join(numericalMAPgenotypes)
        #print site_info
        vcfgenotypeoutput="\t".join(genotypedata)
        output=site_info+ "\t"+vcfgenotypeoutput
        vcfh.write(output+"\n")