X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bam2bcf_indel.c;h=ab9e83ca8f712ae7d7975d67b94b10c702e64908;hb=ccb7838cb53bf0ca6917a77f7991d940057c12db;hp=3a8782b9e473b2fdc76c779468cfa98547d1caa8;hpb=7d93d105e56d3bcee4be728077716794dddf21e7;p=samtools.git diff --git a/bam2bcf_indel.c b/bam2bcf_indel.c index 3a8782b..ab9e83c 100644 --- a/bam2bcf_indel.c +++ b/bam2bcf_indel.c @@ -3,12 +3,14 @@ #include #include "bam.h" #include "bam2bcf.h" -#include "ksort.h" #include "kaln.h" #include "kprobaln.h" #include "khash.h" KHASH_SET_INIT_STR(rg) +#include "ksort.h" +KSORT_INIT_GENERIC(uint32_t) + #define MINUS_CONST 0x10000000 #define INDEL_WINDOW_SIZE 50 @@ -64,7 +66,7 @@ static int tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, for (k = 0; k < c->n_cigar; ++k) { int op = cigar[k] & BAM_CIGAR_MASK; int l = cigar[k] >> BAM_CIGAR_SHIFT; - if (op == BAM_CMATCH) { + if (op == BAM_CMATCH || op == BAM_CEQUAL || op == BAM_CDIFF) { if (c->pos > tpos) return y; if (x + l > tpos) { *_tpos = tpos; @@ -110,7 +112,6 @@ static inline int est_indelreg(int pos, const char *ref, int l, char *ins4) int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_callaux_t *bca, const char *ref, const void *rghash) { - extern void ks_introsort_uint32_t(int, uint32_t*); int i, s, j, k, t, n_types, *types, max_rd_len, left, right, max_ins, *score1, *score2, max_ref2; int N, K, l_run, ref_type, n_alt; char *inscns = 0, *ref2, *query, **ref_sample; @@ -160,6 +161,11 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla if (j > max_rd_len) max_rd_len = j; } } + // To prevent long stretches of N's to be mistaken for indels (sometimes thousands of bases), + // check the number of N's in the sequence and skip places where half or more reference bases are Ns. + int nN=0; for (i=pos; i-posi ) return -1; + ks_introsort(uint32_t, m, aux); // squeeze out identical types for (i = 1, n_types = 1; i < m; ++i) @@ -167,6 +173,12 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla if (n_types == 1 || (double)n_alt / n_tot < bca->min_frac || n_alt < bca->min_support) { // then skip free(aux); return -1; } + if (n_types >= 64) { + free(aux); + if (bam_verbose >= 2) + fprintf(stderr, "[%s] excessive INDEL alleles at position %d. Skip the position.\n", __func__, pos + 1); + return -1; + } types = (int*)calloc(n_types, sizeof(int)); t = 0; types[t++] = aux[0] - MINUS_CONST; @@ -177,7 +189,6 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla for (t = 0; t < n_types; ++t) if (types[t] == 0) break; ref_type = t; // the index of the reference type (0) - assert(n_types < 64); } { // calculate left and right boundary left = pos > INDEL_WINDOW_SIZE? pos - INDEL_WINDOW_SIZE : 0; @@ -195,8 +206,8 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla * sequence realignment helps to increase the power. */ { // construct per-sample consensus - int L = right - left + 1; - uint32_t *cns; + int L = right - left + 1, max_i, max2_i; + uint32_t *cns, max, max2; char *ref0, *r; ref_sample = calloc(n, sizeof(void*)); cns = calloc(L, 4); @@ -216,7 +227,7 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla for (k = 0; k < b->core.n_cigar; ++k) { int op = cigar[k]&0xf; int j, l = cigar[k]>>4; - if (op == BAM_CMATCH) { + if (op == BAM_CMATCH || op == BAM_CEQUAL || op == BAM_CDIFF) { for (j = 0; j < l; ++j) if (x + j >= left && x + j < right) cns[x+j-left] += (bam1_seqi(seq, y+j) == ref0[x+j-left])? 1 : 0x10000; @@ -226,8 +237,16 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla } } // determine the consensus - for (i = 0; i < right - left; ++i) - r[i] = (cns[i] && (double)(cns[i]&0xffff) / ((cns[i]&0xffff)+(cns[i]>>16&0xffff)) < 0.7)? 15 : ref0[i]; + for (i = 0; i < right - left; ++i) r[i] = ref0[i]; + max = max2 = 0; max_i = max2_i = -1; + for (i = 0; i < right - left; ++i) { + if (cns[i]>>16 >= max>>16) max2 = max, max2_i = max_i, max = cns[i], max_i = i; + else if (cns[i]>>16 >= max2>>16) max2 = cns[i], max2_i = i; + } + if ((double)(max&0xffff) / ((max&0xffff) + (max>>16)) >= 0.7) max_i = -1; + if ((double)(max2&0xffff) / ((max2&0xffff) + (max2>>16)) >= 0.7) max2_i = -1; + if (max_i >= 0) r[max_i] = 15; + if (max2_i >= 0) r[max2_i] = 15; // for (i = 0; i < right - left; ++i) fputc("=ACMGRSVTWYHKDBN"[(int)r[i]], stderr); fputc('\n', stderr); } free(ref0); free(cns); @@ -310,8 +329,14 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla // align each read to ref2 for (i = 0; i < n_plp[s]; ++i, ++K) { bam_pileup1_t *p = plp[s] + i; - int qbeg, qend, tbeg, tend, sc; + int qbeg, qend, tbeg, tend, sc, kk; uint8_t *seq = bam1_seq(p->b); + uint32_t *cigar = bam1_cigar(p->b); + if (p->b->core.flag&4) continue; // unmapped reads + // FIXME: the following loop should be better moved outside; nonetheless, realignment should be much slower anyway. + for (kk = 0; kk < p->b->core.n_cigar; ++kk) + if ((cigar[kk]&BAM_CIGAR_MASK) == BAM_CREF_SKIP) break; + if (kk < p->b->core.n_cigar) continue; // FIXME: the following skips soft clips, but using them may be more sensitive. // determine the start and end of sequences for alignment qbeg = tpos2qpos(&p->b->core, bam1_cigar(p->b), left, 0, &tbeg); @@ -406,9 +431,11 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla indelQ2 = tmp > 111? 0 : (int)((1. - tmp/111.) * indelQ2 + .499); // pick the smaller between indelQ1 and indelQ2 indelQ = indelQ1 < indelQ2? indelQ1 : indelQ2; - p->aux = (sc[0]&0x3f)<<16 | seqQ<<8 | indelQ; + if (indelQ > 255) indelQ = 255; + if (seqQ > 255) seqQ = 255; + p->aux = (sc[0]&0x3f)<<16 | seqQ<<8 | indelQ; // use 22 bits in total sumq[sc[0]&0x3f] += indelQ < seqQ? indelQ : seqQ; -// fprintf(stderr, "pos=%d read=%d:%d name=%s call=%d q=%d\n", pos, s, i, bam1_qname(p->b), types[sc[0]&0x3f], indelQ); +// fprintf(stderr, "pos=%d read=%d:%d name=%s call=%d indelQ=%d seqQ=%d\n", pos, s, i, bam1_qname(p->b), types[sc[0]&0x3f], indelQ, seqQ); } } // determine bca->indel_types[] and bca->inscns