X-Git-Url: https://git.donarmstrong.com/?p=rsem.git;a=blobdiff_plain;f=sam%2Fbam2bcf.c;fp=sam%2Fbam2bcf.c;h=340b10b8ff2ebfc1807af7f2c3b30b35a3dade3e;hp=dec3305340f689a100d29dc4f9b5998e7692c6d4;hb=dbcf1cfb8ad1086c21d64e249f012809403e7ddc;hpb=58d504aaf36ae486b1dba6d03e0e9f1c25855037 diff --git a/sam/bam2bcf.c b/sam/bam2bcf.c index dec3305..340b10b 100644 --- a/sam/bam2bcf.c +++ b/sam/bam2bcf.c @@ -1,5 +1,6 @@ #include #include +#include #include "bam.h" #include "kstring.h" #include "bam2bcf.h" @@ -26,20 +27,55 @@ bcf_callaux_t *bcf_call_init(double theta, int min_baseQ) bca->e = errmod_init(1. - theta); bca->min_frac = 0.002; bca->min_support = 1; - return bca; + bca->per_sample_flt = 0; + bca->npos = 100; + bca->ref_pos = calloc(bca->npos, sizeof(int)); + bca->alt_pos = calloc(bca->npos, sizeof(int)); + return bca; +} + + +static int get_position(const bam_pileup1_t *p, int *len) +{ + int icig, n_tot_bases = 0, iread = 0, edist = p->qpos + 1; + for (icig=0; icigb->core.n_cigar; icig++) + { + // Conversion from uint32_t to MIDNSHP + // 0123456 + // MIDNSHP + int cig = bam1_cigar(p->b)[icig] & BAM_CIGAR_MASK; + int ncig = bam1_cigar(p->b)[icig] >> BAM_CIGAR_SHIFT; + if ( cig==0 ) + { + n_tot_bases += ncig; + iread += ncig; + } + else if ( cig==1 ) + { + n_tot_bases += ncig; + iread += ncig; + } + else if ( cig==4 ) + { + iread += ncig; + if ( iread<=p->qpos ) edist -= ncig; + } + } + *len = n_tot_bases; + return edist; } void bcf_call_destroy(bcf_callaux_t *bca) { if (bca == 0) return; errmod_destroy(bca->e); + if (bca->npos) { free(bca->ref_pos); free(bca->alt_pos); bca->npos = 0; } free(bca->bases); free(bca->inscns); free(bca); } /* ref_base is the 4-bit representation of the reference base. It is * negative if we are looking at an indel. */ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t *bca, bcf_callret1_t *r) { - static int *var_pos = NULL, nvar_pos = 0; int i, n, ref4, is_indel, ori_depth = 0; memset(r, 0, sizeof(bcf_callret1_t)); if (ref_base >= 0) { @@ -54,8 +90,7 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t bca->bases = (uint16_t*)realloc(bca->bases, 2 * bca->max_bases); } // fill the bases array - memset(r, 0, sizeof(bcf_callret1_t)); - for (i = n = 0; i < _n; ++i) { + for (i = n = r->n_supp = 0; i < _n; ++i) { const bam_pileup1_t *p = pl + i; int q, b, mapQ, baseQ, is_diff, min_dist, seqQ; // set base @@ -78,6 +113,7 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t b = p->aux>>16&0x3f; is_diff = (b != 0); } + if (is_diff) ++r->n_supp; bca->bases[n++] = q<<5 | (int)bam1_strand(p->b)<<4 | b; // collect annotations if (b < 4) r->qsum[b] += q; @@ -91,97 +127,163 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t r->anno[2<<2|is_diff<<1|1] += mapQ * mapQ; r->anno[3<<2|is_diff<<1|0] += min_dist; r->anno[3<<2|is_diff<<1|1] += min_dist * min_dist; + + // collect read positions for ReadPosBias + int len, pos = get_position(p, &len); + int epos = (double)pos/(len+1) * bca->npos; + if ( bam1_seqi(bam1_seq(p->b),p->qpos) == ref_base ) + bca->ref_pos[epos]++; + else + bca->alt_pos[epos]++; } r->depth = n; r->ori_depth = ori_depth; // glfgen errmod_cal(bca->e, n, 5, bca->bases, r->p); + return r->depth; +} - // Calculate the Variant Distance Bias (make it optional?) - if ( nvar_pos < _n ) { - nvar_pos = _n; - var_pos = realloc(var_pos,sizeof(int)*nvar_pos); - } - int alt_dp=0, read_len=0; - for (i=0; i<_n; i++) { - const bam_pileup1_t *p = pl + i; - if ( bam1_seqi(bam1_seq(p->b),p->qpos) == ref_base ) - continue; +double mann_whitney_1947(int n, int m, int U) +{ + if (U<0) return 0; + if (n==0||m==0) return U==0 ? 1 : 0; + return (double)n/(n+m)*mann_whitney_1947(n-1,m,U-m) + (double)m/(n+m)*mann_whitney_1947(n,m-1,U); +} - var_pos[alt_dp] = p->qpos; - if ( (bam1_cigar(p->b)[0]&BAM_CIGAR_MASK)==4 ) - var_pos[alt_dp] -= bam1_cigar(p->b)[0]>>BAM_CIGAR_SHIFT; +void calc_ReadPosBias(bcf_callaux_t *bca, bcf_call_t *call) +{ + int i, nref = 0, nalt = 0; + unsigned long int U = 0; + for (i=0; inpos; i++) + { + nref += bca->ref_pos[i]; + nalt += bca->alt_pos[i]; + U += nref*bca->alt_pos[i]; + bca->ref_pos[i] = 0; + bca->alt_pos[i] = 0; + } +#if 0 +//todo + double var = 0, avg = (double)(nref+nalt)/bca->npos; + for (i=0; inpos; i++) + { + double ediff = bca->ref_pos[i] + bca->alt_pos[i] - avg; + var += ediff*ediff; + bca->ref_pos[i] = 0; + bca->alt_pos[i] = 0; + } + call->read_pos.avg = avg; + call->read_pos.var = sqrt(var/bca->npos); + call->read_pos.dp = nref+nalt; +#endif + if ( !nref || !nalt ) + { + call->read_pos_bias = -1; + return; + } - alt_dp++; - read_len += p->b->core.l_qseq; + if ( nref>=8 || nalt>=8 ) + { + // normal approximation + double mean = ((double)nref*nalt+1.0)/2.0; + double var2 = (double)nref*nalt*(nref+nalt+1.0)/12.0; + double z = (U-mean)/sqrt(var2); + call->read_pos_bias = z; + //fprintf(stderr,"nref=%d nalt=%d U=%ld mean=%e var=%e zval=%e\n", nref,nalt,U,mean,sqrt(var2),call->read_pos_bias); } - float mvd=0; - int j; - n=0; - for (i=0; i= 1./sqrt(var2*2*M_PI) ) z = 0; // equal to mean + else + { + if ( U >= nref*nalt/2. ) z = sqrt(-2*log(sqrt(var2*2*M_PI)*p)); + else z = -sqrt(-2*log(sqrt(var2*2*M_PI)*p)); } + call->read_pos_bias = z; + //fprintf(stderr,"nref=%d nalt=%d U=%ld p=%e var2=%e zval=%e\n", nref,nalt,U, p,var2,call->read_pos_bias); } - r->mvd[0] = n ? mvd/n : 0; - r->mvd[1] = alt_dp; - r->mvd[2] = alt_dp ? read_len/alt_dp : 0; - - return r->depth; } - -void calc_vdb(int n, const bcf_callret1_t *calls, bcf_call_t *call) +float mean_diff_to_prob(float mdiff, int dp, int readlen) { - // Variant distance bias. Samples merged by means of DP-weighted average. - - float weight=0, tot_prob=0; - - int i; - for (i=0; i2*mu ? 0 : sin(mvd*3.14/2/mu) / (4*mu/3.14); - } - else - { - // Scaled gaussian curve, crude approximation, but behaves well. Using fixed depth for bigger depths. - if ( dp>5 ) - dp = 5; - float sigma2 = (read_len/1.9/(dp+1)) * (read_len/1.9/(dp+1)); - float norm = 1.125*sqrt(2*3.14*sigma2); - float mu = read_len/2.9; - if ( mvd < mu ) - prob = exp(-(mvd-mu)*(mvd-mu)/2/sigma2)/norm; - else - prob = exp(-(mvd-mu)*(mvd-mu)/3.125/sigma2)/norm; - } + float m, v; + if ( dp>=24 ) + { + m = readlen/8.; + if (dp>100) dp = 100; + v = 1.476/(0.182*pow(dp,0.514)); + v = v*(readlen/100.); + } + else + { + m = mv[dp][0]; + v = mv[dp][1]; + m = m*readlen/100.; + v = v*readlen/100.; + v *= 1.2; // allow more variability + } + return 1.0/(v*sqrt(2*M_PI)) * exp(-0.5*((mdiff-m)/v)*((mdiff-m)/v)); +} - //fprintf(stderr,"dp=%d mvd=%d read_len=%d -> prob=%f\n", dp,mvd,read_len,prob); - tot_prob += prob*dp; - weight += dp; +void calc_vdb(bcf_callaux_t *bca, bcf_call_t *call) +{ + int i, dp = 0; + float mean_pos = 0, mean_diff = 0; + for (i=0; inpos; i++) + { + if ( !bca->alt_pos[i] ) continue; + dp += bca->alt_pos[i]; + int j = inpos/2 ? i : bca->npos - i; + mean_pos += bca->alt_pos[i]*j; + } + if ( dp<2 ) + { + call->vdb = -1; + return; } - tot_prob = weight ? tot_prob/weight : 1; - //fprintf(stderr,"prob=%f\n", tot_prob); - call->vdb = tot_prob; + mean_pos /= dp; + for (i=0; inpos; i++) + { + if ( !bca->alt_pos[i] ) continue; + int j = inpos/2 ? i : bca->npos - i; + mean_diff += bca->alt_pos[i] * fabs(j - mean_pos); + } + mean_diff /= dp; + call->vdb = mean_diff_to_prob(mean_diff, dp, bca->npos); } -int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, bcf_call_t *call) +/** + * bcf_call_combine() - sets the PL array and VDB, RPB annotations, finds the top two alleles + * @n: number of samples + * @calls: each sample's calls + * @bca: auxiliary data structure for holding temporary values + * @ref_base: the reference base + * @call: filled with the annotations + */ +int bcf_call_combine(int n, const bcf_callret1_t *calls, bcf_callaux_t *bca, int ref_base /*4-bit*/, bcf_call_t *call) { int ref4, i, j, qsum[4]; int64_t tmp; @@ -194,6 +296,8 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, for (i = 0; i < n; ++i) for (j = 0; j < 4; ++j) qsum[j] += calls[i].qsum[j]; + int qsum_tot=0; + for (j=0; j<4; j++) { qsum_tot += qsum[j]; call->qsum[j] = 0; } for (j = 0; j < 4; ++j) qsum[j] = qsum[j] << 2 | j; // find the top 2 alleles for (i = 1; i < 4; ++i) // insertion sort @@ -205,9 +309,15 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, call->a[0] = ref4; for (i = 3, j = 1; i >= 0; --i) { if ((qsum[i]&3) != ref4) { - if (qsum[i]>>2 != 0) call->a[j++] = qsum[i]&3; + if (qsum[i]>>2 != 0) + { + if ( j<4 ) call->qsum[j] = (float)(qsum[i]>>2)/qsum_tot; // ref N can make j>=4 + call->a[j++] = qsum[i]&3; + } else break; } + else + call->qsum[0] = (float)(qsum[i]>>2)/qsum_tot; } if (ref_base >= 0) { // for SNPs, find the "unseen" base if (((ref4 < 4 && j < 4) || (ref4 == 4 && j < 5)) && i >= 0) @@ -255,12 +365,13 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, for (j = 0; j < 16; ++j) call->anno[j] += calls[i].anno[j]; } - calc_vdb(n, calls, call); + calc_vdb(bca, call); + calc_ReadPosBias(bca, call); return 0; } -int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bcr, int is_SP, +int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bcr, int fmt_flag, const bcf_callaux_t *bca, const char *ref) { extern double kt_fisher_exact(int n11, int n12, int n21, int n22, double *_left, double *_right, double *two); @@ -304,34 +415,39 @@ int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bc } kputc('\0', &s); // INFO - if (bc->ori_ref < 0) kputs("INDEL;", &s); + if (bc->ori_ref < 0) ksprintf(&s,"INDEL;IS=%d,%f;", bca->max_support, bca->max_frac); kputs("DP=", &s); kputw(bc->ori_depth, &s); kputs(";I16=", &s); for (i = 0; i < 16; ++i) { if (i) kputc(',', &s); kputw(bc->anno[i], &s); } - if ( bc->vdb!=1 ) - { - ksprintf(&s, ";VDB=%.4f", bc->vdb); - } + //ksprintf(&s,";RPS=%d,%f,%f", bc->read_pos.dp,bc->read_pos.avg,bc->read_pos.var); + ksprintf(&s,";QS=%f,%f,%f,%f", bc->qsum[0],bc->qsum[1],bc->qsum[2],bc->qsum[3]); + if (bc->vdb != -1) + ksprintf(&s, ";VDB=%e", bc->vdb); + if (bc->read_pos_bias != -1 ) + ksprintf(&s, ";RPB=%e", bc->read_pos_bias); kputc('\0', &s); // FMT kputs("PL", &s); - if (bcr) { - kputs(":DP", &s); - if (is_SP) kputs(":SP", &s); + if (bcr && fmt_flag) { + if (fmt_flag & B2B_FMT_DP) kputs(":DP", &s); + if (fmt_flag & B2B_FMT_DV) kputs(":DV", &s); + if (fmt_flag & B2B_FMT_SP) kputs(":SP", &s); } kputc('\0', &s); b->m_str = s.m; b->str = s.s; b->l_str = s.l; bcf_sync(b); memcpy(b->gi[0].data, bc->PL, b->gi[0].len * bc->n); - if (bcr) { - uint16_t *dp = (uint16_t*)b->gi[1].data; - int32_t *sp = is_SP? b->gi[2].data : 0; + if (bcr && fmt_flag) { + uint16_t *dp = (fmt_flag & B2B_FMT_DP)? b->gi[1].data : 0; + uint16_t *dv = (fmt_flag & B2B_FMT_DV)? b->gi[1 + ((fmt_flag & B2B_FMT_DP) != 0)].data : 0; + int32_t *sp = (fmt_flag & B2B_FMT_SP)? b->gi[1 + ((fmt_flag & B2B_FMT_DP) != 0) + ((fmt_flag & B2B_FMT_DV) != 0)].data : 0; for (i = 0; i < bc->n; ++i) { bcf_callret1_t *p = bcr + i; - dp[i] = p->depth < 0xffff? p->depth : 0xffff; - if (is_SP) { + if (dp) dp[i] = p->depth < 0xffff? p->depth : 0xffff; + if (dv) dv[i] = p->n_supp < 0xffff? p->n_supp : 0xffff; + if (sp) { if (p->anno[0] + p->anno[1] < 2 || p->anno[2] + p->anno[3] < 2 || p->anno[0] + p->anno[2] < 2 || p->anno[1] + p->anno[3] < 2) {