X-Git-Url: http://woldlab.caltech.edu/gitweb/?p=samtools.git;a=blobdiff_plain;f=bam2bcf.c;h=dec3305340f689a100d29dc4f9b5998e7692c6d4;hp=e55212ca98cbb28041fb6a6dfc46619ff595ee09;hb=0f906dafb2ad22371a753557562ef95c3034480d;hpb=8d2494d1fb7cd0fa7c63be5ffba8dd1a11457522 diff --git a/bam2bcf.c b/bam2bcf.c index e55212c..dec3305 100644 --- a/bam2bcf.c +++ b/bam2bcf.c @@ -11,23 +11,21 @@ extern void ks_introsort_uint32_t(size_t n, uint32_t a[]); #define CALL_ETA 0.03f #define CALL_MAX 256 #define CALL_DEFTHETA 0.83f +#define DEF_MAPQ 20 #define CAP_DIST 25 -struct __bcf_callaux_t { - int max_bases, capQ, min_baseQ; - uint16_t *bases; - errmod_t *e; -}; - bcf_callaux_t *bcf_call_init(double theta, int min_baseQ) { bcf_callaux_t *bca; if (theta <= 0.) theta = CALL_DEFTHETA; bca = calloc(1, sizeof(bcf_callaux_t)); bca->capQ = 60; + bca->openQ = 40; bca->extQ = 20; bca->tandemQ = 100; bca->min_baseQ = min_baseQ; bca->e = errmod_init(1. - theta); + bca->min_frac = 0.002; + bca->min_support = 1; return bca; } @@ -35,16 +33,20 @@ void bcf_call_destroy(bcf_callaux_t *bca) { if (bca == 0) return; errmod_destroy(bca->e); - free(bca->bases); free(bca); + free(bca->bases); free(bca->inscns); free(bca); } - -int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base /*4-bit*/, bcf_callaux_t *bca, bcf_callret1_t *r) +/* 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) { - int i, n, ref4; + static int *var_pos = NULL, nvar_pos = 0; + int i, n, ref4, is_indel, ori_depth = 0; memset(r, 0, sizeof(bcf_callret1_t)); - ref4 = bam_nt16_nt4_table[ref_base]; + if (ref_base >= 0) { + ref4 = bam_nt16_nt4_table[ref_base]; + is_indel = 0; + } else ref4 = 4, is_indel = 1; if (_n == 0) return -1; - // enlarge the bases array if necessary if (bca->max_bases < _n) { bca->max_bases = _n; @@ -55,21 +57,30 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base /*4-bit*/, bcf memset(r, 0, sizeof(bcf_callret1_t)); for (i = n = 0; i < _n; ++i) { const bam_pileup1_t *p = pl + i; - int q, b, mapQ, baseQ, is_diff, min_dist; + int q, b, mapQ, baseQ, is_diff, min_dist, seqQ; // set base - if (p->is_del || (p->b->core.flag&BAM_FUNMAP)) continue; // skip unmapped reads and deleted bases - baseQ = q = (int)bam1_qual(p->b)[p->qpos]; // base quality + if (p->is_del || p->is_refskip || (p->b->core.flag&BAM_FUNMAP)) continue; + ++ori_depth; + baseQ = q = is_indel? p->aux&0xff : (int)bam1_qual(p->b)[p->qpos]; // base/indel quality + seqQ = is_indel? (p->aux>>8&0xff) : 99; if (q < bca->min_baseQ) continue; - mapQ = p->b->core.qual < bca->capQ? p->b->core.qual : bca->capQ; + if (q > seqQ) q = seqQ; + mapQ = p->b->core.qual < 255? p->b->core.qual : DEF_MAPQ; // special case for mapQ==255 + mapQ = mapQ < bca->capQ? mapQ : bca->capQ; if (q > mapQ) q = mapQ; if (q > 63) q = 63; if (q < 4) q = 4; - b = bam1_seqi(bam1_seq(p->b), p->qpos); // base - b = bam_nt16_nt4_table[b? b : ref_base]; // b is the 2-bit base + if (!is_indel) { + b = bam1_seqi(bam1_seq(p->b), p->qpos); // base + b = bam_nt16_nt4_table[b? b : ref_base]; // b is the 2-bit base + is_diff = (ref4 < 4 && b == ref4)? 0 : 1; + } else { + b = p->aux>>16&0x3f; + is_diff = (b != 0); + } bca->bases[n++] = q<<5 | (int)bam1_strand(p->b)<<4 | b; // collect annotations - r->qsum[b] += q; - is_diff = (ref4 < 4 && b == ref4)? 0 : 1; + if (b < 4) r->qsum[b] += q; ++r->anno[0<<2|is_diff<<1|bam1_strand(p->b)]; min_dist = p->b->core.l_qseq - 1 - p->qpos; if (min_dist > p->qpos) min_dist = p->qpos; @@ -81,18 +92,103 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base /*4-bit*/, bcf r->anno[3<<2|is_diff<<1|0] += min_dist; r->anno[3<<2|is_diff<<1|1] += min_dist * min_dist; } - r->depth = n; + r->depth = n; r->ori_depth = ori_depth; // glfgen errmod_cal(bca->e, n, 5, bca->bases, r->p); + + // 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; + + 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; + + alt_dp++; + read_len += p->b->core.l_qseq; + } + float mvd=0; + int j; + n=0; + for (i=0; imvd[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) +{ + // 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; + } + + //fprintf(stderr,"dp=%d mvd=%d read_len=%d -> prob=%f\n", dp,mvd,read_len,prob); + tot_prob += prob*dp; + weight += dp; + } + tot_prob = weight ? tot_prob/weight : 1; + //fprintf(stderr,"prob=%f\n", tot_prob); + call->vdb = tot_prob; +} + int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, bcf_call_t *call) { int ref4, i, j, qsum[4]; int64_t tmp; - call->ori_ref = ref4 = bam_nt16_nt4_table[ref_base]; - if (ref4 > 4) ref4 = 4; + if (ref_base >= 0) { + call->ori_ref = ref4 = bam_nt16_nt4_table[ref_base]; + if (ref4 > 4) ref4 = 4; + } else call->ori_ref = -1, ref4 = 0; // calculate qsum memset(qsum, 0, 4 * sizeof(int)); for (i = 0; i < n; ++i) @@ -113,9 +209,14 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, else break; } } - if (((ref4 < 4 && j < 4) || (ref4 == 4 && j < 5)) && i >= 0) - call->unseen = j, call->a[j++] = qsum[i]&3; - call->n_alleles = j; + if (ref_base >= 0) { // for SNPs, find the "unseen" base + if (((ref4 < 4 && j < 4) || (ref4 == 4 && j < 5)) && i >= 0) + call->unseen = j, call->a[j++] = qsum[i]&3; + call->n_alleles = j; + } else { + call->n_alleles = j; + if (call->n_alleles == 1) return -1; // no reliable supporting read. stop doing anything + } // set the PL array if (call->n < n) { call->n = n; @@ -127,8 +228,8 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, x = call->n_alleles * (call->n_alleles + 1) / 2; // get the possible genotypes for (i = z = 0; i < call->n_alleles; ++i) - for (j = i; j < call->n_alleles; ++j) - g[z++] = call->a[i] * 5 + call->a[j]; + for (j = 0; j <= i; ++j) + g[z++] = call->a[j] * 5 + call->a[i]; for (i = 0; i < n; ++i) { uint8_t *PL = call->PL + x * i; const bcf_callret1_t *r = calls + i; @@ -143,44 +244,108 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, PL[j] = y; } } +// if (ref_base < 0) fprintf(stderr, "%d,%d,%f,%d\n", call->n_alleles, x, sum_min, call->unseen); call->shift = (int)(sum_min + .499); } // combine annotations memset(call->anno, 0, 16 * sizeof(int)); - for (i = call->depth = 0, tmp = 0; i < n; ++i) { + for (i = call->depth = call->ori_depth = 0, tmp = 0; i < n; ++i) { call->depth += calls[i].depth; + call->ori_depth += calls[i].ori_depth; for (j = 0; j < 16; ++j) call->anno[j] += calls[i].anno[j]; } + + calc_vdb(n, calls, call); + return 0; } -int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b) +int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bcr, int is_SP, + 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); kstring_t s; - int i; + int i, j; b->n_smpl = bc->n; b->tid = tid; b->pos = pos; b->qual = 0; s.s = b->str; s.m = b->m_str; s.l = 0; kputc('\0', &s); - kputc("ACGTN"[bc->ori_ref], &s); kputc('\0', &s); - for (i = 1; i < 5; ++i) { - if (bc->a[i] < 0) break; - if (i > 1) kputc(',', &s); - kputc(bc->unseen == i? 'X' : "ACGT"[bc->a[i]], &s); + if (bc->ori_ref < 0) { // an indel + // write REF + kputc(ref[pos], &s); + for (j = 0; j < bca->indelreg; ++j) kputc(ref[pos+1+j], &s); + kputc('\0', &s); + // write ALT + kputc(ref[pos], &s); + for (i = 1; i < 4; ++i) { + if (bc->a[i] < 0) break; + if (i > 1) { + kputc(',', &s); kputc(ref[pos], &s); + } + if (bca->indel_types[bc->a[i]] < 0) { // deletion + for (j = -bca->indel_types[bc->a[i]]; j < bca->indelreg; ++j) + kputc(ref[pos+1+j], &s); + } else { // insertion; cannot be a reference unless a bug + char *inscns = &bca->inscns[bc->a[i] * bca->maxins]; + for (j = 0; j < bca->indel_types[bc->a[i]]; ++j) + kputc("ACGTN"[(int)inscns[j]], &s); + for (j = 0; j < bca->indelreg; ++j) kputc(ref[pos+1+j], &s); + } + } + kputc('\0', &s); + } else { // a SNP + kputc("ACGTN"[bc->ori_ref], &s); kputc('\0', &s); + for (i = 1; i < 5; ++i) { + if (bc->a[i] < 0) break; + if (i > 1) kputc(',', &s); + kputc(bc->unseen == i? 'X' : "ACGT"[bc->a[i]], &s); + } + kputc('\0', &s); } kputc('\0', &s); - kputc('\0', &s); // INFO - kputs("I16=", &s); + if (bc->ori_ref < 0) kputs("INDEL;", &s); + 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); + } kputc('\0', &s); // FMT - kputs("PL", &s); kputc('\0', &s); + kputs("PL", &s); + if (bcr) { + kputs(":DP", &s); + if (is_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; + 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 (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) + { + sp[i] = 0; + } else { + double left, right, two; + int x; + kt_fisher_exact(p->anno[0], p->anno[1], p->anno[2], p->anno[3], &left, &right, &two); + x = (int)(-4.343 * log(two) + .499); + if (x > 255) x = 255; + sp[i] = x; + } + } + } + } return 0; }