void CHMM::genotypeProbability(vector<SNP*>::iterator snp_it, double probs[2])
{
probs[0] = probs[1] = 1.0;
char ref = (*snp_it)->GetRef();
char alt = (*snp_it)->GetAlt();
int read_count = (*snp_it)->GetReadCount();
double errate = 0.01;
// REVISIT: I may need to take the nth root here before returning the probabilities
// n here is read_count
for(int count=0; count<read_count; count++) {
READ *rd = (*snp_it)->GetRead(count);
char all = rd->GetAllele(count);
int hap = rd->GetHap();
double happrob = rd->GetHapProb();
if(all==ref) {
probs[0] = probs[0];
probs[1] *= (1-happrob);
} else if(all==alt) {
probs[0] *= (1-happrob);
probs[1] = probs[1];
} else {
probs[0] *= errate;
probs[1] *= errate;
}
}
// Experimental: Logs for revisit above
probs[0] = pow(2.7182, log(probs[0])/read_count);
probs[1] = pow(2.7182, log(probs[1])/read_count);
probs[2] = pow(2.7182, log(probs[2])/read_count);
}
void CHMM::FindFBDistance(CObsSeq *obsSeq, ostream &outFile, long snp_start, long snp_end)
{
double *scale;
double **alpha, **beta;
double logProb;
int i, j, em, t, k;
double **posterior;
double post;
for(i=1;i<=obsSeq->mNbSequences;i++){ // Loop over observation files:
long T = obsSeq->mNbObs[i];
CObs **obs = obsSeq->mObs[i];
alpha = SetMatrix(T, mN);// different size every time
beta = SetMatrix(T, mN);
scale = SetVector(T);
int *q = SetIntVector(T);
posterior = SetMatrix(T,mN);
// logProb contains log P(x)
// Haplotype Calling
logProb = ForwardAlgo(alpha, scale, obs, T, TRUE);
BackwardAlgo(beta, scale, obs, T);
for(t=1;t<=T;t++) {
READ *nd = (((CFlexibleObs<READ*>*)(obs[t]))->Get(1));
for(k=1;k<=mN;k++) {
// Experimental: P(x) cancels scaling for alpha and beta
// so posterior is just their product
double st = 1.0;
for(j=T; j>t; j--) {
st *= scale[j];
}
post = (alpha[t][k]*beta[t][k]*scale[t]);
#ifdef DEBUG
cout << endl;
cout << "Scale = " << scale[t] << endl;
cout << "Scaled alpha = " << alpha[t][k] << endl;
cout << "beta = " << beta[t][k] << endl;
cout << "Scaled beta = " << beta[t][k]*scale[t] << endl;
#endif
posterior[t][k] = post;
}
// REVISIT: hardcoding #states.
q[t] = posterior[t][1] > posterior[t][2] ? 1 : 2;
#ifdef DEBUG
//cout << "q[t] = " << q[t] << endl;
//cout << "posterior = " << posterior[t][q[t]] << endl;
#endif
nd->assignHaplotype(q[t],posterior[t][q[t]]);
}
UpdateGenotypes(snp_start, snp_end);
for(j=2;j<=T;j++) {
outFile << q[j] << "\t" << ((reads_list)[reads_list.size()-T+j-1])->GetPos() << endl;
}
delete [] q;
}
}
// Works on known snp list
// Given t-th read in list, returns allele of snp with position pos.
char GetPosAllele(int t, long pos)
{
READ *rd = reads_list[t];
int snp_count = rd->GetKnownCount();
SNP ** snp_list = rd->GetKnownList();
for(int i=0; i<snp_count; i++) {
if(snp_list[i]->GetPos()==pos)
return rd->GetKnownAllele(i);
}
return 'N';
}
//double CHMM::haplotypeProbability(int refct, int altct, int errct, int g)
void CHMM::haplotypeProbability(vector<SNP*>::iterator snp_it, double probs[3])
{
probs[0] = 1.0;
probs[1] = 1.0;
probs[2] = 1.0;
char ref = (*snp_it)->GetRef();
char alt = (*snp_it)->GetAlt();
double qual = (*snp_it)->GetQualScore();
double qualscore = pow(10,-(qual/10)); // P(alt==wrong)
double errate = 0.01;
//cout << "SNP: " << (*snp_it)->GetPos() << " Qual: " << qual << ", " << qualscore << endl;
// Here I assume that if an observed allele is not ref (or alt in case of BB),
// then it belongs to the other haplotype.
for(int count=0; count<(*snp_it)->GetReadCount(); count++) {
READ *rd = (*snp_it)->GetRead(count);
char all = rd->GetAllele(count);
double haprob = rd->GetHapProb();
if(all == ref) {
probs[0] = probs[0]; //Do nothing
probs[1] *= (1-haprob);//*qualscore;
probs[2] *= ((1-haprob)*(qualscore));
} else if(all = alt) {
probs[0] *= ((1-haprob)*(qualscore));
probs[1] *= (1-haprob);//*qualscore;
probs[2] = probs[2]; //Do nothing
} else {
probs[0] *= errate;
probs[1] *= errate;
probs[2] *= errate;
}
#ifdef DEBUG
//cout << "qual = " << qual << endl;
//cout << "qualscore = " << qualscore << endl;
//cout << "haprob = " << haprob << endl;
//cout << "Computing happrobs:" << "\t" << probs[0] << "\t" << probs[1] << "\t" << probs[2] << endl;
#endif
}
}
void NaiveBayes(int snp_start, int snp_end, int T)
{
int i, j, t;
SNP **known_snp_list = new SNP*[100];
int *known_index = new int[200];
int reads_list_size = reads_list.size();
for (t = 1+reads_list_size-T; t <= reads_list_size-1; t++) {
int known_snp_count = 0, flag = 0, direction = 1, hap = 0, mismatchsum = 0;
double norm = 0.0;
double emission_list[3], haprob[3], emission_set[3][100];
READ *pd = reads_list[t-1];
READ *nd = reads_list[t];
int plen = pd->GetLen();
int nlen = nd->GetLen();
int ppos = pd->GetPos();
int npos = nd->GetPos();
int rd_start = ppos < npos ? npos : ppos;
int rd_end = ppos+plen > npos+nlen ? npos+nlen-1 : ppos+plen-1;
//mismatchsum = GetMismatchSum(nd);
//if(mismatchsum>=50)
// continue;
GetKnownSnpList(known_snp_list, &known_snp_count, known_index, t);
// This adds the count of known overlapping snps between adjacent reads
(*nd).AddKnownCount(known_snp_count);
#ifdef DEBUG
if(known_snp_count>0)
cout << "Read " << nd->GetPos() << " ; "<< t << " and " << t+1 << ", " << rd_start << ", " << rd_end << " with " << known_snp_count << " known snps." << endl << endl;
cout << "SnpPos R A LL..\tAA_gen AA_obs AA_genob\tAB_gen AB_obs AB_genob\tBB_gen BB_obs BB_genob\tprob\n";
cout << "Throwing emissions from read: " << nd->GetPos() << endl;
#endif
for(j=1;j<=2;j++) {
#ifdef DEBUG
cout << "Haplotype " << j << endl;
#endif
emission_list[j] = 0.0;
for(int count=0; count<known_snp_count; count++) {
double emission;
SNP *sp = known_snp_list[count];
#ifdef DEBUG
cout << sp->GetPos() << " " << sp->GetRef() << " " << sp->GetAlt();
#endif
// Obtains naive bayes score for the count-th overlapping known snp
// between the adjacent reads for haplotype, 'j'.
// Working with relative haplotype here.
emission = compute_new_emission(known_snp_list, count, t, known_index, j);
if(emission==0.0)
continue;
emission_list[j] += ((emission));
emission_set[j][count] = emission;
}
#ifdef DEBUG
cout << "Total Emission = " << emission_list[j] << endl;
#endif
}
if(emission_set[1][0]<emission_set[2][0])
direction = 2;
for(int count=1; count<known_snp_count; count++) {
if(emission_set[1][count]>emission_set[2][count]&&direction==2)
flag = 1;
if(emission_set[1][count]<emission_set[2][count]&&direction==1)
flag = 1;
}
#ifdef DEBUG
cout << "Discordance = " << flag << endl;
#endif
norm = emission_list[1] + emission_list[2];
if(norm!=0.0) {
haprob[1] = emission_list[1] / norm;
haprob[2] = emission_list[2] / norm;
} else {
haprob[1] = 0.5;
haprob[2] = 0.5;
}
#ifdef DEBUG
cout << "Happrob[1] = " << haprob[1] << endl;
cout << "Happrob[2] = " << haprob[2] << endl;
#endif
// If haplotype probabilities are equal, randomly assign haplotype
hap = haprob[1] > haprob[2] ? 1 : haprob[1] == haprob[2] ? t%2+1 : 2;
double happrob = haprob[hap];
// convert relative haplotype to absolute haplotype
hap = (*pd).GetHap() == hap ? 1 : 2;
nd->assignHaplotype(hap, happrob, flag);
flag = 0; direction = 1;
}
delete [] known_index;
delete [] known_snp_list;
}
// Calculate posterior probability
int somaticHaplotypeProbability(vector<SNP*>::iterator snp_it, double probs[3], int *known_hap_count)
{
char ref = (*snp_it)->GetRef();
char alt = (*snp_it)->GetAlt();
int real_count = 0, ref_ct = 0, alt_ct = 0;
int ref1_ct = 0, alt1_ct = 0, ref2_ct = 0, alt2_ct = 0, err_ct=0;
int contig_max = 0, contig_new = 1, max_stretch = 0;
int count = 0, gap = 0, stretch = 0, contiguous = 1, contig_start = 0;
int mapq1 = 0, proximal_ins_sum = 0, proximal_del_sum = 0;
int clus_flag = 0, clus_in_flag = 1, clus_del_flag = 1;
int max_reads = (*snp_it)->GetReadCount();
double probs1[3], probs2[3];
double prevalence;
ref_ct = (*snp_it)->GetRefCount();
alt_ct = (*snp_it)->GetAltCount();
// dynamically calculate the prevalence
prevalence = (double)(alt_ct)/(double)(ref_ct+alt_ct);
*known_hap_count = 0;
probs[0] = 1.0; probs[1] = 1.0; probs[2] = 1.0;
probs1[0] = 1.0; probs1[1] = 1.0; probs1[2] = 1.0;
probs2[0] = 1.0; probs2[1] = 1.0; probs2[2] = 1.0;
#ifdef DEBUG
cout << " Count on snp " << (*snp_it)->GetPos() << " is " << max_reads << endl;
#endif
// First for loop looks for maximum contiguous region
for(count=0; count<max_reads; count++) {
READ *rd = (*snp_it)->GetRead(count);
READ *pd = (*snp_it)->GetRead(count-1);
if(rd->GetKnownOverlapCount()>0&&rd->GetDiscordance()==0) { // No gap exists in hap assignment
(*known_hap_count)++;
stretch++;
if(contig_new==1) {
contig_start = count;
contig_new = 0;
}
} else { // Gap exists in hap assignment. Not contiguous.
contiguous = 0;
contig_new = 1;
if(stretch>max_stretch) { // Identify longest stretch of contiguous assignments
max_stretch = stretch;
contig_max = contig_start;
}
stretch = 0;
}
if(contiguous==1) {
max_stretch=stretch;
contig_max=0;
} else if(stretch>max_stretch) {
max_stretch = stretch;
contig_max = contig_start;
}
#ifdef DEBUG
cout << "SNP=" << (*snp_it)->GetPos() << " Gap=" << gap << " contiguous=" << contiguous << " max_reads=" << max_reads << " max_stretch=" << max_stretch << " contig_max=" << contig_max << endl;
#endif
}
if(contiguous==0)
gap+=2;
if(max_stretch<(int)(9*max_reads/10)) // If max stretch of contiguous assigned reads is at least 90% of total number, consider reads for genotype score assignments.
gap++;
#ifdef DEBUG
cout << " Gap=" << gap << " contiguous=" << contiguous << " max_reads=" << max_reads << " max_stretch=" << max_stretch << " contig_max=" << contig_max << " prevalence = " << prevalence << endl;
#endif
// Second for loop performs actual genotype score calculation
for(count=0; count<max_reads; count++) {
bool proximal_ins = 0, proximal_del = 0;
char all, pall = 'N';
int flag = 0;
double qual, qualscore;
READ *rd = (*snp_it)->GetRead(count);
READ *pd = (*snp_it)->GetRead(count-1);
int hap = rd->GetHap();
double haprob = rd->GetHapProb();
double hethap = sqrt((1.0-haprob)*haprob);
int snp_count = rd->GetSnpCount();
if(snp_count==0) {
#ifdef DEBUG
cout << "For som " << (*snp_it)->GetPos() << " skipping read " << rd->GetPos() << " for happrob calculation" << endl;
#endif
haprob = 0.5;
}
// Obtain allele, base qual and #proximal insertions/deletions on current read
long snp_position = (*snp_it)->GetPos();
for(int s_pos=0; s_pos<snp_count; s_pos++) {
if(rd->GetSnp(s_pos)->GetPos() == snp_position) {
proximal_ins = rd->GetProximalInsert(s_pos);
proximal_del = rd->GetProximalDelete(s_pos);
all = rd->GetAllele(s_pos);
qual = rd->GetQualScore(s_pos)-33;
qualscore = pow(10.0,-(qual/10.0));
break;
}
}
if(hap==1) { // See comments later
if(all == ref) {
ref1_ct++;
} else if(all == alt) {
alt1_ct++;
} else
err_ct++;
} else if(hap==2) {
if(all == ref) {
ref2_ct++;
} else if(all == alt) {
alt2_ct++;
} else
err_ct++;
}
real_count++;
if(qual<5) // No base with quality < 5 should be considered
continue;
if(all==alt && qual>=20) // At least one alternate allele with base quality >20
mapq1++;
if(proximal_ins==1) { // Read contains insertion in vicinity
proximal_ins_sum+= 1;
}
if(proximal_del==1) { // Read contains deletion in vicinity
proximal_del_sum+= 1;
}
if(all==alt) {
clus_in_flag--;
clus_del_flag--;
if(proximal_ins==1)
clus_in_flag++;
if(proximal_del==1)
clus_del_flag++;
}
proximal_ins = 0;
proximal_del = 0;
// Work on contiguous stretch only
if(count>=contig_max&&count<contig_max+max_stretch) {
// Since we are haplotype-aware, two separate set of calculations are performed
// First set assumes somatic mutation lies on haplotype 1
// Second set assumes somatic mutation lies on haplotype 2
// In the end we consider that haplotype to contain somatic mutation, which has higher het probability assigned
if(hap==1) {
if(all == ref) {
probs1[0] *= ((haprob)*(1.0-qualscore));
probs1[1] *= ((1.0-prevalence-qualscore/4.0)*haprob);
probs1[2] *= ((1.0-haprob) * (qualscore/3.0));
probs2[0] *= ((haprob)*(1.0-qualscore));
probs2[1] *= ((1.0-prevalence-qualscore/4.0)*(1.0-haprob));
probs2[2] *= ((1.0-haprob) * (qualscore/3.0));
} else if(all == alt) {
probs1[0] *= (haprob*(1.0*qualscore/3.0));
probs1[1] *= ((1.0*qualscore/3.0)*(1.0-haprob));
probs1[2] *= ((1.0-haprob)*(1.0*qualscore/3.0));
probs2[0] *= ((1.0-haprob) * (qualscore/3.0));
probs2[1] *= (prevalence*haprob);
probs2[2] *= ((haprob)*(1.0-qualscore));
}
} else if(hap==2) {
if(all == ref) {
probs1[0] *= ((haprob)*(1.0-qualscore));
probs1[1] *= ((1.0-prevalence-qualscore/4.0)*(1.0-haprob));
probs1[2] *= ((1.0-haprob) * (qualscore/3.0));
probs2[0] *= ((haprob)*(1.0-qualscore));
probs2[1] *= ((1.0-prevalence-qualscore/4.0)*haprob);
probs2[2] *= ((1.0-haprob) * (qualscore/3.0));
} else if(all == alt) {
probs1[0] *= ((1.0-haprob) * (qualscore/3.0));
probs1[1] *= (prevalence*haprob);
probs1[2] *= ((haprob)*(1.0-qualscore));
probs2[0] *= (haprob*(1.0*qualscore/3.0));
probs2[1] *= ((1.0*qualscore/3.0)*(1.0-haprob));
probs2[2] *= ((1.0-haprob)*(1.0*qualscore/3.0));
}
}
} else if(rd->GetKnownOverlapCount()>=0) { // Insufficient contiguous regions. Haplotype unaware
if(all == ref) {
probs1[0] *= ((1.0-qualscore));
probs1[1] *= (1.0-prevalence-qualscore/3.0);
probs1[2] *= ((qualscore/3.0));
probs2[0] *= ((1.0-qualscore));
probs2[1] *= (1.0-prevalence-qualscore/3.0);
probs2[2] *= ((qualscore/3.0));
} else if(all == alt) {
probs1[0] *= ((qualscore/3.0));
probs1[1] *= (prevalence-qualscore/3.0);
probs1[2] *= ((1.0-qualscore));
probs2[0] *= ((qualscore/3.0));
probs2[1] *= (prevalence-qualscore/3.0);
probs2[2] *= ((1.0-qualscore));
} else {
probs1[0] *= ((2.0*qualscore/3.0));
probs1[1] *= (qualscore/2.0);
probs1[2] *= ((2.0*qualscore/3.0));
probs2[0] *= ((2.0*qualscore/3.0));
probs2[1] *= (qualscore/2.0);
probs2[2] *= ((2.0*qualscore/3.0));
}
}
#ifdef DEBUG
cout << "For som " << (*snp_it)->GetPos() << " bearing allele " << all << ref << alt << " on read " << rd->GetPos() << ", qualscore: " << qualscore << " has known_hap_count " << rd->GetKnownOverlapCount() << " and flag " << flag << " bearing haplotype " << hap << " with haprob " << haprob << " for happrob calculation " << probs1[0] << ":" << probs1[1] << ":" << probs1[2] << " and " << probs2[0] << ":" << probs2[1] << ":" << probs2[2] << " and proxdel " << proximal_del << " and proxins " << proximal_ins << " and clus_in " << clus_in_flag << " and clus_del " << clus_del_flag << " and clus " << clus_flag << endl;
#endif
}
// Pick one of two sets of scores
double norm1 = probs1[0] + probs1[1] + probs1[2];
double norm2 = probs2[0] + probs2[1] + probs2[2];
if(probs1[1]/norm1>probs2[1]/norm1) {
probs[0]=probs1[0];probs[1]=probs1[1];probs[2]=probs1[2];
} else if(probs2[1]/norm2>probs1[1]/norm1) {
probs[0]=probs2[0];probs[1]=probs2[1];probs[2]=probs2[2];
} else if(probs1[0]/norm1<probs2[0]/norm2) {
probs[0]=probs1[0];probs[1]=probs1[1];probs[2]=probs1[2];
} else if(probs2[0]/norm2<probs1[0]/norm1) {
probs[0]=probs2[0];probs[1]=probs2[1];probs[2]=probs2[2];
} else {
probs[0]=probs1[0];probs[1]=probs1[1];probs[2]=probs1[2];
}
PROXIMAL_READ_CT = 2 + (int)ceil((double)(real_count)/10.0);
if(clus_in_flag>=1||clus_del_flag>=1)
clus_flag = 1;
// FILTERS
if(ref_ct<3 || alt_ct<1 || (gap!=3&&alt1_ct>=2&&alt2_ct>=2) || mapq1==0 || proximal_ins_sum>=PROXIMAL_READ_CT&&alt_ct<5 || proximal_del_sum>=PROXIMAL_READ_CT&&alt_ct<5 || clus_flag==1) { // || (gap==3&&alt_ct>=ref_ct))
gap=5;
if(mapq1==0)
gap = 6;
if(proximal_ins_sum>=PROXIMAL_READ_CT&&alt_ct<5 || proximal_del_sum>=PROXIMAL_READ_CT&&alt_ct<5)
gap = 7;
if(clus_flag==1)
gap = 8;
probs[0] = 1.0; probs[1] = probs[2] = 0.0;
}
if(probs[0]<0.0||probs[1]<0.0||probs[2]<0.0) {
cout << "Negative: " << probs[0] << ":" << probs[1] << ":" << probs[2] << endl;
probs[0] = 1.0; probs[1] = probs[2] = 0.0;
gap=5;
}
#ifdef DEBUG
cout << "Accepted mutation " << (*snp_it)->GetPos() << " Prox=" << PROXIMAL_READ_CT << " Gap=" << gap << " known=" << (*snp_it)->GetKnown() << " real_count=" << real_count << " ref_ct=" << ref_ct << " ref1_ct=" << ref1_ct << " ref2_ct=" << ref2_ct << " alt_ct=" << alt_ct << " alt1_ct=" << alt1_ct << " alt2_ct=" << alt2_ct << " proxdelsum=" << proximal_del_sum << " proxinssum=" << proximal_ins_sum << " clus_in=" << clus_in_flag << " clus_del=" << clus_del_flag << " clus=" << clus_flag << endl;
#endif
return gap;
}