13 static inline int printw(int c, FILE *fp)
17 if (c == 0) return fputc('0', fp);
18 for (l = 0, x = c < 0? -c : c; x > 0; x /= 10) buf[l++] = x%10 + '0';
19 if (c < 0) buf[l++] = '-';
21 for (x = 0; x < l/2; ++x) {
22 int y = buf[x]; buf[x] = buf[l-1-x]; buf[l-1-x] = y;
28 static inline void pileup_seq(const bam_pileup1_t *p, int pos, int ref_len, const char *ref)
33 putchar(p->b->core.qual > 93? 126 : p->b->core.qual + 33);
36 int c = bam_nt16_rev_table[bam1_seqi(bam1_seq(p->b), p->qpos)];
38 int rb = pos < ref_len? ref[pos] : 'N';
39 if (c == '=' || bam_nt16_table[c] == bam_nt16_table[rb]) c = bam1_strand(p->b)? ',' : '.';
40 else c = bam1_strand(p->b)? tolower(c) : toupper(c);
42 if (c == '=') c = bam1_strand(p->b)? ',' : '.';
43 else c = bam1_strand(p->b)? tolower(c) : toupper(c);
46 } else putchar(p->is_refskip? (bam1_strand(p->b)? '<' : '>') : '*');
48 putchar('+'); printw(p->indel, stdout);
49 for (j = 1; j <= p->indel; ++j) {
50 int c = bam_nt16_rev_table[bam1_seqi(bam1_seq(p->b), p->qpos + j)];
51 putchar(bam1_strand(p->b)? tolower(c) : toupper(c));
53 } else if (p->indel < 0) {
54 printw(p->indel, stdout);
55 for (j = 1; j <= -p->indel; ++j) {
56 int c = (ref && (int)pos+j < ref_len)? ref[pos+j] : 'N';
57 putchar(bam1_strand(p->b)? tolower(c) : toupper(c));
60 if (p->is_tail) putchar('$');
68 #define MPLP_NO_COMP 0x20
69 #define MPLP_NO_ORPHAN 0x40
70 #define MPLP_REALN 0x80
71 #define MPLP_FMT_DP 0x100
72 #define MPLP_FMT_SP 0x200
73 #define MPLP_NO_INDEL 0x400
74 #define MPLP_EXT_BAQ 0x800
75 #define MPLP_ILLUMINA13 0x1000
76 #define MPLP_IGNORE_RG 0x2000
77 #define MPLP_PRINT_POS 0x4000
78 #define MPLP_PRINT_MAPQ 0x8000
80 void *bed_read(const char *fn);
81 void bed_destroy(void *_h);
82 int bed_overlap(const void *_h, const char *chr, int beg, int end);
85 int max_mq, min_mq, flag, min_baseQ, capQ_thres, max_depth, max_indel_depth;
86 int openQ, extQ, tandemQ, min_support; // for indels
87 double min_frac; // for indels
99 const mplp_conf_t *conf;
108 static int mplp_func(void *data, bam1_t *b)
110 extern int bam_realn(bam1_t *b, const char *ref);
111 extern int bam_prob_realn_core(bam1_t *b, const char *ref, int);
112 extern int bam_cap_mapQ(bam1_t *b, char *ref, int thres);
113 mplp_aux_t *ma = (mplp_aux_t*)data;
117 ret = ma->iter? bam_iter_read(ma->fp, ma->iter, b) : bam_read1(ma->fp, b);
119 if (b->core.tid < 0 || (b->core.flag&BAM_FUNMAP)) { // exclude unmapped reads
123 if (ma->conf->bed) { // test overlap
124 skip = !bed_overlap(ma->conf->bed, ma->h->target_name[b->core.tid], b->core.pos, bam_calend(&b->core, bam1_cigar(b)));
127 if (ma->conf->rghash) { // exclude read groups
128 uint8_t *rg = bam_aux_get(b, "RG");
129 skip = (rg && bcf_str2id(ma->conf->rghash, (const char*)(rg+1)) >= 0);
132 if (ma->conf->flag & MPLP_ILLUMINA13) {
134 uint8_t *qual = bam1_qual(b);
135 for (i = 0; i < b->core.l_qseq; ++i)
136 qual[i] = qual[i] > 31? qual[i] - 31 : 0;
138 has_ref = (ma->ref && ma->ref_id == b->core.tid)? 1 : 0;
140 if (has_ref && (ma->conf->flag&MPLP_REALN)) bam_prob_realn_core(b, ma->ref, (ma->conf->flag & MPLP_EXT_BAQ)? 3 : 1);
141 if (has_ref && ma->conf->capQ_thres > 10) {
142 int q = bam_cap_mapQ(b, ma->ref, ma->conf->capQ_thres);
144 else if (b->core.qual > q) b->core.qual = q;
146 else if (b->core.qual < ma->conf->min_mq) skip = 1;
147 else if ((ma->conf->flag&MPLP_NO_ORPHAN) && (b->core.flag&1) && !(b->core.flag&2)) skip = 1;
152 static void group_smpl(mplp_pileup_t *m, bam_sample_t *sm, kstring_t *buf,
153 int n, char *const*fn, int *n_plp, const bam_pileup1_t **plp, int ignore_rg)
156 memset(m->n_plp, 0, m->n * sizeof(int));
157 for (i = 0; i < n; ++i) {
158 for (j = 0; j < n_plp[i]; ++j) {
159 const bam_pileup1_t *p = plp[i] + j;
162 q = ignore_rg? 0 : bam_aux_get(p->b, "RG");
163 if (q) id = bam_smpl_rg2smid(sm, fn[i], (char*)q+1, buf);
164 if (id < 0) id = bam_smpl_rg2smid(sm, fn[i], 0, buf);
165 if (id < 0 || id >= m->n) {
166 assert(q); // otherwise a bug
167 fprintf(pysamerr, "[%s] Read group %s used in file %s but absent from the header or an alignment missing read group.\n", __func__, (char*)q+1, fn[i]);
170 if (m->n_plp[id] == m->m_plp[id]) {
171 m->m_plp[id] = m->m_plp[id]? m->m_plp[id]<<1 : 8;
172 m->plp[id] = realloc(m->plp[id], sizeof(bam_pileup1_t) * m->m_plp[id]);
174 m->plp[id][m->n_plp[id]++] = *p;
179 static int mpileup(mplp_conf_t *conf, int n, char **fn)
181 extern void *bcf_call_add_rg(void *rghash, const char *hdtext, const char *list);
182 extern void bcf_call_del_rghash(void *rghash);
184 int i, tid, pos, *n_plp, tid0 = -1, beg0 = 0, end0 = 1u<<29, ref_len, ref_tid = -1, max_depth, max_indel_depth;
185 const bam_pileup1_t **plp;
191 bcf_callaux_t *bca = 0;
192 bcf_callret1_t *bcr = 0;
197 bam_sample_t *sm = 0;
201 memset(&gplp, 0, sizeof(mplp_pileup_t));
202 memset(&buf, 0, sizeof(kstring_t));
203 memset(&bc, 0, sizeof(bcf_call_t));
204 data = calloc(n, sizeof(void*));
205 plp = calloc(n, sizeof(void*));
206 n_plp = calloc(n, sizeof(int*));
207 sm = bam_smpl_init();
209 // read the header and initialize data
210 for (i = 0; i < n; ++i) {
212 data[i] = calloc(1, sizeof(mplp_aux_t));
213 data[i]->fp = strcmp(fn[i], "-") == 0? bam_dopen(fileno(stdin), "r") : bam_open(fn[i], "r");
214 data[i]->conf = conf;
215 h_tmp = bam_header_read(data[i]->fp);
216 data[i]->h = i? h : h_tmp; // for i==0, "h" has not been set yet
217 bam_smpl_add(sm, fn[i], (conf->flag&MPLP_IGNORE_RG)? 0 : h_tmp->text);
218 rghash = bcf_call_add_rg(rghash, h_tmp->text, conf->pl_list);
222 idx = bam_index_load(fn[i]);
224 fprintf(pysamerr, "[%s] fail to load index for %d-th input.\n", __func__, i+1);
227 if (bam_parse_region(h_tmp, conf->reg, &tid, &beg, &end) < 0) {
228 fprintf(pysamerr, "[%s] malformatted region or wrong seqname for %d-th input.\n", __func__, i+1);
231 if (i == 0) tid0 = tid, beg0 = beg, end0 = end;
232 data[i]->iter = bam_iter_query(idx, tid, beg, end);
233 bam_index_destroy(idx);
235 if (i == 0) h = h_tmp;
237 // FIXME: to check consistency
238 bam_header_destroy(h_tmp);
242 gplp.n_plp = calloc(sm->n, sizeof(int));
243 gplp.m_plp = calloc(sm->n, sizeof(int));
244 gplp.plp = calloc(sm->n, sizeof(void*));
246 fprintf(pysamerr, "[%s] %d samples in %d input files\n", __func__, sm->n, n);
247 // write the VCF header
248 if (conf->flag & MPLP_GLF) {
250 bh = calloc(1, sizeof(bcf_hdr_t));
251 s.l = s.m = 0; s.s = 0;
252 bp = bcf_open("-", (conf->flag&MPLP_NO_COMP)? "wu" : "w");
253 for (i = 0; i < h->n_targets; ++i) {
254 kputs(h->target_name[i], &s);
258 bh->name = malloc(s.l);
259 memcpy(bh->name, s.s, s.l);
261 for (i = 0; i < sm->n; ++i) {
262 kputs(sm->smpl[i], &s); kputc('\0', &s);
265 bh->sname = malloc(s.l);
266 memcpy(bh->sname, s.s, s.l);
267 bh->txt = malloc(strlen(BAM_VERSION) + 64);
268 bh->l_txt = 1 + sprintf(bh->txt, "##samtoolsVersion=%s\n", BAM_VERSION);
271 bcf_hdr_write(bp, bh);
272 bca = bcf_call_init(-1., conf->min_baseQ);
273 bcr = calloc(sm->n, sizeof(bcf_callret1_t));
274 bca->rghash = rghash;
275 bca->openQ = conf->openQ, bca->extQ = conf->extQ, bca->tandemQ = conf->tandemQ;
276 bca->min_frac = conf->min_frac;
277 bca->min_support = conf->min_support;
279 if (tid0 >= 0 && conf->fai) { // region is set
280 ref = faidx_fetch_seq(conf->fai, h->target_name[tid0], 0, 0x7fffffff, &ref_len);
282 for (i = 0; i < n; ++i) data[i]->ref = ref, data[i]->ref_id = tid0;
283 } else ref_tid = -1, ref = 0;
284 iter = bam_mplp_init(n, mplp_func, (void**)data);
285 max_depth = conf->max_depth;
286 if (max_depth * sm->n > 1<<20)
287 fprintf(pysamerr, "(%s) Max depth is above 1M. Potential memory hog!\n", __func__);
288 if (max_depth * sm->n < 8000) {
289 max_depth = 8000 / sm->n;
290 fprintf(pysamerr, "<%s> Set max per-file depth to %d\n", __func__, max_depth);
292 max_indel_depth = conf->max_indel_depth * sm->n;
293 bam_mplp_set_maxcnt(iter, max_depth);
294 while (bam_mplp_auto(iter, &tid, &pos, n_plp, plp) > 0) {
295 if (conf->reg && (pos < beg0 || pos >= end0)) continue; // out of the region requested
296 if (conf->bed && tid >= 0 && !bed_overlap(conf->bed, h->target_name[tid], pos, pos+1)) continue;
297 if (tid != ref_tid) {
299 if (conf->fai) ref = faidx_fetch_seq(conf->fai, h->target_name[tid], 0, 0x7fffffff, &ref_len);
300 for (i = 0; i < n; ++i) data[i]->ref = ref, data[i]->ref_id = tid;
303 if (conf->flag & MPLP_GLF) {
304 int total_depth, _ref0, ref16;
305 bcf1_t *b = calloc(1, sizeof(bcf1_t));
306 for (i = total_depth = 0; i < n; ++i) total_depth += n_plp[i];
307 group_smpl(&gplp, sm, &buf, n, fn, n_plp, plp, conf->flag & MPLP_IGNORE_RG);
308 _ref0 = (ref && pos < ref_len)? ref[pos] : 'N';
309 ref16 = bam_nt16_table[_ref0];
310 for (i = 0; i < gplp.n; ++i)
311 bcf_call_glfgen(gplp.n_plp[i], gplp.plp[i], ref16, bca, bcr + i);
312 bcf_call_combine(gplp.n, bcr, ref16, &bc);
313 bcf_call2bcf(tid, pos, &bc, b, (conf->flag&(MPLP_FMT_DP|MPLP_FMT_SP))? bcr : 0,
314 (conf->flag&MPLP_FMT_SP), 0, 0);
315 bcf_write(bp, bh, b);
318 if (!(conf->flag&MPLP_NO_INDEL) && total_depth < max_indel_depth && bcf_call_gap_prep(gplp.n, gplp.n_plp, gplp.plp, pos, bca, ref, rghash) >= 0) {
319 for (i = 0; i < gplp.n; ++i)
320 bcf_call_glfgen(gplp.n_plp[i], gplp.plp[i], -1, bca, bcr + i);
321 if (bcf_call_combine(gplp.n, bcr, -1, &bc) >= 0) {
322 b = calloc(1, sizeof(bcf1_t));
323 bcf_call2bcf(tid, pos, &bc, b, (conf->flag&(MPLP_FMT_DP|MPLP_FMT_SP))? bcr : 0,
324 (conf->flag&MPLP_FMT_SP), bca, ref);
325 bcf_write(bp, bh, b);
330 printf("%s\t%d\t%c", h->target_name[tid], pos + 1, (ref && pos < ref_len)? ref[pos] : 'N');
331 for (i = 0; i < n; ++i) {
333 printf("\t%d\t", n_plp[i]);
335 printf("*\t*"); // FIXME: printf() is very slow...
336 if (conf->flag & MPLP_PRINT_POS) printf("\t*");
338 for (j = 0; j < n_plp[i]; ++j)
339 pileup_seq(plp[i] + j, pos, ref_len, ref);
341 for (j = 0; j < n_plp[i]; ++j) {
342 const bam_pileup1_t *p = plp[i] + j;
343 int c = bam1_qual(p->b)[p->qpos] + 33;
344 if (c > 126) c = 126;
347 if (conf->flag & MPLP_PRINT_MAPQ) {
349 for (j = 0; j < n_plp[i]; ++j) {
350 int c = plp[i][j].b->core.qual + 33;
351 if (c > 126) c = 126;
355 if (conf->flag & MPLP_PRINT_POS) {
357 for (j = 0; j < n_plp[i]; ++j) {
358 if (j > 0) putchar(',');
359 printf("%d", plp[i][j].qpos + 1); // FIXME: printf() is very slow...
369 bam_smpl_destroy(sm); free(buf.s);
370 for (i = 0; i < gplp.n; ++i) free(gplp.plp[i]);
371 free(gplp.plp); free(gplp.n_plp); free(gplp.m_plp);
372 bcf_call_del_rghash(rghash);
373 bcf_hdr_destroy(bh); bcf_call_destroy(bca); free(bc.PL); free(bcr);
374 bam_mplp_destroy(iter);
375 bam_header_destroy(h);
376 for (i = 0; i < n; ++i) {
377 bam_close(data[i]->fp);
378 if (data[i]->iter) bam_iter_destroy(data[i]->iter);
381 free(data); free(plp); free(ref); free(n_plp);
385 #define MAX_PATH_LEN 1024
386 static int read_file_list(const char *file_list,int *n,char **argv[])
388 char buf[MAX_PATH_LEN];
392 FILE *fh = fopen(file_list,"r");
395 fprintf(pysamerr,"%s: %s\n", file_list,strerror(errno));
399 // Speed is not an issue here, determine the number of files by reading the file twice
401 while ( fgets(buf,MAX_PATH_LEN,fh) ) nfiles++;
403 if ( fseek(fh, 0L, SEEK_SET) )
405 fprintf(pysamerr,"%s: %s\n", file_list,strerror(errno));
409 files = calloc(nfiles,sizeof(char*));
411 while ( fgets(buf,MAX_PATH_LEN,fh) )
414 while ( len>0 && isspace(buf[len-1]) ) len--;
415 if ( !len ) continue;
417 files[nfiles] = malloc(sizeof(char)*(len+1));
418 strncpy(files[nfiles],buf,len);
419 files[nfiles][len] = 0;
425 fprintf(pysamerr,"No files read from %s\n", file_list);
434 int bam_mpileup(int argc, char *argv[])
437 const char *file_list = NULL;
439 int nfiles = 0, use_orphan = 0;
441 memset(&mplp, 0, sizeof(mplp_conf_t));
442 #define MPLP_PRINT_POS 0x4000
446 mplp.max_depth = 250; mplp.max_indel_depth = 250;
447 mplp.openQ = 40; mplp.extQ = 20; mplp.tandemQ = 100;
448 mplp.min_frac = 0.002; mplp.min_support = 1;
449 mplp.flag = MPLP_NO_ORPHAN | MPLP_REALN;
450 while ((c = getopt(argc, argv, "Agf:r:l:M:q:Q:uaRC:BDSd:L:b:P:o:e:h:Im:F:EG:6Os")) >= 0) {
453 mplp.fai = fai_load(optarg);
454 if (mplp.fai == 0) return 1;
456 case 'd': mplp.max_depth = atoi(optarg); break;
457 case 'r': mplp.reg = strdup(optarg); break;
458 case 'l': mplp.bed = bed_read(optarg); break;
459 case 'P': mplp.pl_list = strdup(optarg); break;
460 case 'g': mplp.flag |= MPLP_GLF; break;
461 case 'u': mplp.flag |= MPLP_NO_COMP | MPLP_GLF; break;
462 case 'a': mplp.flag |= MPLP_NO_ORPHAN | MPLP_REALN; break;
463 case 'B': mplp.flag &= ~MPLP_REALN; break;
464 case 'D': mplp.flag |= MPLP_FMT_DP; break;
465 case 'S': mplp.flag |= MPLP_FMT_SP; break;
466 case 'I': mplp.flag |= MPLP_NO_INDEL; break;
467 case 'E': mplp.flag |= MPLP_EXT_BAQ; break;
468 case '6': mplp.flag |= MPLP_ILLUMINA13; break;
469 case 'R': mplp.flag |= MPLP_IGNORE_RG; break;
470 case 's': mplp.flag |= MPLP_PRINT_MAPQ; break;
471 case 'O': mplp.flag |= MPLP_PRINT_POS; break;
472 case 'C': mplp.capQ_thres = atoi(optarg); break;
473 case 'M': mplp.max_mq = atoi(optarg); break;
474 case 'q': mplp.min_mq = atoi(optarg); break;
475 case 'Q': mplp.min_baseQ = atoi(optarg); break;
476 case 'b': file_list = optarg; break;
477 case 'o': mplp.openQ = atoi(optarg); break;
478 case 'e': mplp.extQ = atoi(optarg); break;
479 case 'h': mplp.tandemQ = atoi(optarg); break;
480 case 'A': use_orphan = 1; break;
481 case 'F': mplp.min_frac = atof(optarg); break;
482 case 'm': mplp.min_support = atoi(optarg); break;
483 case 'L': mplp.max_indel_depth = atoi(optarg); break;
487 mplp.rghash = bcf_str2id_init();
488 if ((fp_rg = fopen(optarg, "r")) == 0)
489 fprintf(pysamerr, "(%s) Fail to open file %s. Continue anyway.\n", __func__, optarg);
490 while (!feof(fp_rg) && fscanf(fp_rg, "%s", buf) > 0) // this is not a good style, but forgive me...
491 bcf_str2id_add(mplp.rghash, strdup(buf));
497 if (use_orphan) mplp.flag &= ~MPLP_NO_ORPHAN;
499 fprintf(pysamerr, "\n");
500 fprintf(pysamerr, "Usage: samtools mpileup [options] in1.bam [in2.bam [...]]\n\n");
501 fprintf(pysamerr, "Input options:\n\n");
502 fprintf(pysamerr, " -6 assume the quality is in the Illumina-1.3+ encoding\n");
503 fprintf(pysamerr, " -A count anomalous read pairs\n");
504 fprintf(pysamerr, " -B disable BAQ computation\n");
505 fprintf(pysamerr, " -b FILE list of input BAM files [null]\n");
506 fprintf(pysamerr, " -C INT parameter for adjusting mapQ; 0 to disable [0]\n");
507 fprintf(pysamerr, " -d INT max per-BAM depth to avoid excessive memory usage [%d]\n", mplp.max_depth);
508 fprintf(pysamerr, " -E extended BAQ for higher sensitivity but lower specificity\n");
509 fprintf(pysamerr, " -f FILE faidx indexed reference sequence file [null]\n");
510 fprintf(pysamerr, " -G FILE exclude read groups listed in FILE [null]\n");
511 fprintf(pysamerr, " -l FILE list of positions (chr pos) or regions (BED) [null]\n");
512 fprintf(pysamerr, " -M INT cap mapping quality at INT [%d]\n", mplp.max_mq);
513 fprintf(pysamerr, " -r STR region in which pileup is generated [null]\n");
514 fprintf(pysamerr, " -R ignore RG tags\n");
515 fprintf(pysamerr, " -q INT skip alignments with mapQ smaller than INT [%d]\n", mplp.min_mq);
516 fprintf(pysamerr, " -Q INT skip bases with baseQ/BAQ smaller than INT [%d]\n", mplp.min_baseQ);
517 fprintf(pysamerr, "\nOutput options:\n\n");
518 fprintf(pysamerr, " -D output per-sample DP in BCF (require -g/-u)\n");
519 fprintf(pysamerr, " -g generate BCF output (genotype likelihoods)\n");
520 fprintf(pysamerr, " -O output base positions on reads (disabled by -g/-u)\n");
521 fprintf(pysamerr, " -s output mapping quality (disabled by -g/-u)\n");
522 fprintf(pysamerr, " -S output per-sample strand bias P-value in BCF (require -g/-u)\n");
523 fprintf(pysamerr, " -u generate uncompress BCF output\n");
524 fprintf(pysamerr, "\nSNP/INDEL genotype likelihoods options (effective with `-g' or `-u'):\n\n");
525 fprintf(pysamerr, " -e INT Phred-scaled gap extension seq error probability [%d]\n", mplp.extQ);
526 fprintf(pysamerr, " -F FLOAT minimum fraction of gapped reads for candidates [%g]\n", mplp.min_frac);
527 fprintf(pysamerr, " -h INT coefficient for homopolymer errors [%d]\n", mplp.tandemQ);
528 fprintf(pysamerr, " -I do not perform indel calling\n");
529 fprintf(pysamerr, " -L INT max per-sample depth for INDEL calling [%d]\n", mplp.max_indel_depth);
530 fprintf(pysamerr, " -m INT minimum gapped reads for indel candidates [%d]\n", mplp.min_support);
531 fprintf(pysamerr, " -o INT Phred-scaled gap open sequencing error probability [%d]\n", mplp.openQ);
532 fprintf(pysamerr, " -P STR comma separated list of platforms for indels [all]\n");
533 fprintf(pysamerr, "\n");
534 fprintf(pysamerr, "Notes: Assuming diploid individuals.\n\n");
538 if ( read_file_list(file_list,&nfiles,&fn) ) return 1;
539 mpileup(&mplp,nfiles,fn);
540 for (c=0; c<nfiles; c++) free(fn[c]);
542 } else mpileup(&mplp, argc - optind, argv + optind);
543 if (mplp.rghash) bcf_str2id_thorough_destroy(mplp.rghash);
544 free(mplp.reg); free(mplp.pl_list);
545 if (mplp.fai) fai_destroy(mplp.fai);
546 if (mplp.bed) bed_destroy(mplp.bed);