X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bam2bcf_indel.c;h=c28b6859dd0101dc5f9ded34a42ded1add58f781;hb=1a1147a5aef19776be078f3be4bd4cc7b9aa8ceb;hp=7359cc8855cf56d8e4c3255c23a52726401f72b7;hpb=0bb4884ffb1f4c9485efee763dd63b575f2c1381;p=samtools.git diff --git a/bam2bcf_indel.c b/bam2bcf_indel.c index 7359cc8..c28b685 100644 --- a/bam2bcf_indel.c +++ b/bam2bcf_indel.c @@ -1,14 +1,60 @@ #include +#include +#include #include "bam.h" #include "bam2bcf.h" #include "ksort.h" #include "kaln.h" - -#define INDEL_DEBUG +#include "khash.h" +KHASH_SET_INIT_STR(rg) #define MINUS_CONST 0x10000000 #define INDEL_WINDOW_SIZE 50 -#define INDEL_BAD_SCORE 10000 + +void *bcf_call_add_rg(void *_hash, const char *hdtext, const char *list) +{ + const char *s, *p, *q, *r, *t; + khash_t(rg) *hash; + if (list == 0 || hdtext == 0) return _hash; + if (_hash == 0) _hash = kh_init(rg); + hash = (khash_t(rg)*)_hash; + if ((s = strstr(hdtext, "@RG\t")) == 0) return hash; + do { + t = strstr(s + 4, "@RG\t"); // the next @RG + if ((p = strstr(s, "\tID:")) != 0) p += 4; + if ((q = strstr(s, "\tPL:")) != 0) q += 4; + if (p && q && (t == 0 || (p < t && q < t))) { // ID and PL are both present + int lp, lq; + char *x; + for (r = p; *r && *r != '\t' && *r != '\n'; ++r); lp = r - p; + for (r = q; *r && *r != '\t' && *r != '\n'; ++r); lq = r - q; + x = calloc((lp > lq? lp : lq) + 1, 1); + for (r = q; *r && *r != '\t' && *r != '\n'; ++r) x[r-q] = *r; + if (strstr(list, x)) { // insert ID to the hash table + khint_t k; + int ret; + for (r = p; *r && *r != '\t' && *r != '\n'; ++r) x[r-p] = *r; + x[r-p] = 0; + k = kh_get(rg, hash, x); + if (k == kh_end(hash)) k = kh_put(rg, hash, x, &ret); + else free(x); + } else free(x); + } + s = t; + } while (s); + return hash; +} + +void bcf_call_del_rghash(void *_hash) +{ + khint_t k; + khash_t(rg) *hash = (khash_t(rg)*)_hash; + if (hash == 0) return; + for (k = kh_begin(hash); k < kh_end(hash); ++k) + if (kh_exist(hash, k)) + free((char*)kh_key(hash, k)); + kh_destroy(rg, hash); +} static int tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, int is_left, int32_t *_tpos) { @@ -37,39 +83,71 @@ static int tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, *_tpos = x; return last_y; } +// FIXME: check if the inserted sequence is consistent with the homopolymer run // l is the relative gap length and l_run is the length of the homopolymer on the reference static inline int est_seqQ(const bcf_callaux_t *bca, int l, int l_run) { int q, qh; q = bca->openQ + bca->extQ * (abs(l) - 1); - qh = l_run >= 3? (int)(bca->tandemQ * (double)l / l_run + .499) : 1000; + qh = l_run >= 3? (int)(bca->tandemQ * (double)abs(l) / l_run + .499) : 1000; return q < qh? q : qh; } -int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_callaux_t *bca, const char *ref) +static inline int est_indelreg(int pos, const char *ref, int l, char *ins4) +{ + int i, j, max = 0, max_i = pos, score = 0; + l = abs(l); + for (i = pos + 1, j = 0; ref[i]; ++i, ++j) { + if (ins4) score += (toupper(ref[i]) != "ACGTN"[(int)ins4[j%l]])? -10 : 1; + else score += (toupper(ref[i]) != toupper(ref[pos+1+j%l]))? -10 : 1; + if (score < 0) break; + if (max < score) max = score, max_i = i; + } + return max_i - pos; +} + +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, *score, N, K, l_run, ref_type; + int i, s, j, k, t, n_types, *types, max_rd_len, left, right, max_ins, *score, N, K, l_run, ref_type, n_alt; char *inscns = 0, *ref2, *query; + khash_t(rg) *hash = (khash_t(rg)*)rghash; if (ref == 0 || bca == 0) return -1; + // mark filtered reads + if (rghash) { + N = 0; + for (s = N = 0; s < n; ++s) { + for (i = 0; i < n_plp[s]; ++i) { + bam_pileup1_t *p = plp[s] + i; + const uint8_t *rg = bam_aux_get(p->b, "RG"); + p->aux = 1; // filtered by default + if (rg) { + khint_t k = kh_get(rg, hash, (const char*)(rg + 1)); + if (k != kh_end(hash)) p->aux = 0, ++N; // not filtered + } + } + } + if (N == 0) return -1; // no reads left + } // determine if there is a gap - for (s = 0; s < n; ++s) { + for (s = N = 0; s < n; ++s) { for (i = 0; i < n_plp[s]; ++i) if (plp[s][i].indel != 0) break; if (i < n_plp[s]) break; } if (s == n) return -1; // there is no indel at this position. + for (s = N = 0; s < n; ++s) N += n_plp[s]; // N is the total number of reads { // find out how many types of indels are present int m; uint32_t *aux; - aux = calloc(n + 1, 4); + aux = calloc(N + 1, 4); m = max_rd_len = 0; aux[m++] = MINUS_CONST; // zero indel is always a type - for (s = N = 0; s < n; ++s) { - N += n_plp[s]; // N is the total number of reads + for (s = 0; s < n; ++s) { for (i = 0; i < n_plp[s]; ++i) { const bam_pileup1_t *p = plp[s] + i; - if (p->indel != 0) + if (p->indel != 0 && (rghash == 0 || p->aux == 0)) aux[m++] = MINUS_CONST + p->indel; j = bam_cigar2qlen(&p->b->core, bam1_cigar(p->b)); if (j > max_rd_len) max_rd_len = j; @@ -79,7 +157,9 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla // squeeze out identical types for (i = 1, n_types = 1; i < m; ++i) if (aux[i] != aux[i-1]) ++n_types; - assert(n_types > 1); // there must at least one type of non-reference indel + if (n_types == 1) { // no indels + free(aux); return -1; + } types = (int*)calloc(n_types, sizeof(int)); t = 0; types[t++] = aux[0] - MINUS_CONST; @@ -151,10 +231,17 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla ref2 = calloc(right - left + max_ins + 2, 1); query = calloc(right - left + max_rd_len + max_ins + 2, 1); score = calloc(N * n_types, sizeof(int)); + bca->indelreg = 0; for (t = 0; t < n_types; ++t) { - int l; + int l, ir; ka_param2_t ap = ka_param2_qual; ap.band_width = abs(types[t]) + 3; + // compute indelreg + if (types[t] == 0) ir = 0; + else if (types[t] > 0) ir = est_indelreg(pos, ref, types[t], &inscns[t*max_ins]); + else ir = est_indelreg(pos, ref, -types[t], 0); + if (ir > bca->indelreg) bca->indelreg = ir; +// fprintf(stderr, "%d, %d, %d\n", pos, types[t], ir); // write ref2 for (k = 0, j = left; j <= pos; ++j) ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]]; @@ -178,30 +265,29 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla // determine the start and end of sequences for alignment qbeg = tpos2qpos(&p->b->core, bam1_cigar(p->b), left, 0, &tbeg); qend = tpos2qpos(&p->b->core, bam1_cigar(p->b), right, 1, &tend); - assert(tbeg >= left); + if (types[t] < 0) { + int l = -types[t]; + tbeg = tbeg - l > left? tbeg - l : left; + } // write the query sequence for (l = qbeg; l < qend; ++l) query[l - qbeg] = bam_nt16_nt4_table[bam1_seqi(seq, l)]; // do alignment; this takes most of computing time for indel calling -#ifdef INDEL_DEBUG - for (sc = 0; sc < tend - tbeg + types[t]; ++sc) - fputc("ACGTN"[(int)ref2[tbeg-left+sc]], stderr); - fputc('\n', stderr); - for (sc = 0; sc < qend - qbeg; ++sc) fputc("ACGTN"[(int)query[qbeg + sc]], stderr); - fputc('\n', stderr); -#endif sc = ka_global_score((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]), - (uint8_t*)query + qbeg, qend - qbeg, &ap); -#ifdef INDEL_DEBUG - fprintf(stderr, "pos=%d type=%d read=%d:%d name=%s score=%d\n", pos, types[t], s, i, bam1_qname(p->b), sc); -#endif + (uint8_t*)query, qend - qbeg, &ap); score[K*n_types + t] = -sc; +/* + for (l = 0; l < tend - tbeg + abs(types[t]); ++l) + fputc("ACGTN"[(int)ref2[tbeg-left+l]], stderr); + fputc('\n', stderr); + for (l = 0; l < qend - qbeg; ++l) fputc("ACGTN"[(int)query[l]], stderr); + fputc('\n', stderr); + fprintf(stderr, "pos=%d type=%d read=%d:%d name=%s qbeg=%d tbeg=%d score=%d\n", pos, types[t], s, i, bam1_qname(p->b), qbeg, tbeg, sc); +*/ } } } free(ref2); free(query); - { // choose the top 4 indel types; reference must be included - } { // compute indelQ int *sc, tmp, *sumq; sc = alloca(n_types * sizeof(int)); @@ -210,7 +296,7 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla for (s = K = 0; s < n; ++s) { for (i = 0; i < n_plp[s]; ++i, ++K) { bam_pileup1_t *p = plp[s] + i; - int *sct = &score[K*n_types], indelQ; + int *sct = &score[K*n_types], indelQ, seqQ; for (t = 0; t < n_types; ++t) sc[t] = sct[t]<<6 | t; for (t = 1; t < n_types; ++t) // insertion sort for (j = t; j > 0 && sc[j] < sc[j-1]; --j) @@ -223,21 +309,21 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla */ if ((sc[0]&0x3f) == ref_type) { indelQ = (sc[1]>>6) - (sc[0]>>6); - tmp = est_seqQ(bca, types[sc[1]&0x3f], l_run); + seqQ = est_seqQ(bca, types[sc[1]&0x3f], l_run); } else { for (t = 0; t < n_types; ++t) // look for the reference type if ((sc[t]&0x3f) == ref_type) break; indelQ = (sc[t]>>6) - (sc[0]>>6); - tmp = est_seqQ(bca, types[sc[0]&0x3f], l_run); + seqQ = est_seqQ(bca, types[sc[0]&0x3f], l_run); } - if (indelQ > tmp) indelQ = tmp; - if (indelQ > p->b->core.qual) indelQ = p->b->core.qual; - if (indelQ > bca->capQ) indelQ = bca->capQ; - p->aux = (sc[0]&0x3f)<<8 | indelQ; - sumq[sc[0]&0x3f] += indelQ; + p->aux = (sc[0]&0x3f)<<16 | seqQ<<8 | indelQ; + 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); } } - // determine bca->indel_types[] + // determine bca->indel_types[] and bca->inscns + bca->maxins = max_ins; + bca->inscns = realloc(bca->inscns, bca->maxins * 4); for (t = 0; t < n_types; ++t) sumq[t] = sumq[t]<<6 | t; for (t = 1; t < n_types; ++t) // insertion sort @@ -251,12 +337,25 @@ int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_calla sumq[0] = tmp; } for (t = 0; t < 4; ++t) bca->indel_types[t] = B2B_INDEL_NULL; - for (t = 0; t < 4 && t < n_types; ++t) + for (t = 0; t < 4 && t < n_types; ++t) { bca->indel_types[t] = types[sumq[t]&0x3f]; + memcpy(&bca->inscns[t * bca->maxins], &inscns[(sumq[t]&0x3f) * max_ins], bca->maxins); + } + // update p->aux + for (s = n_alt = 0; s < n; ++s) { + for (i = 0; i < n_plp[s]; ++i) { + bam_pileup1_t *p = plp[s] + i; + int x = types[p->aux>>16&0x3f]; + for (j = 0; j < 4; ++j) + if (x == bca->indel_types[j]) break; + p->aux = j<<16 | (j == 4? 0 : (p->aux&0xffff)); + if ((p->aux>>16&0x3f) > 0) ++n_alt; +// fprintf(stderr, "pos=%d read=%d:%d name=%s call=%d q=%d\n", pos, s, i, bam1_qname(p->b), p->aux>>16&63, p->aux&0xff); + } + } } - // FIXME: to set the inserted sequence free(score); // free free(types); free(inscns); - return 0; + return n_alt > 0? 0 : -1; }