+
+/***********
+ * mpileup *
+ ***********/
+
+#include <assert.h>
+#include "bam2bcf.h"
+#include "sample.h"
+
+#define MPLP_GLF 0x10
+#define MPLP_NO_COMP 0x20
+#define MPLP_NO_ORPHAN 0x40
+#define MPLP_REALN 0x80
+#define MPLP_FMT_DP 0x100
+#define MPLP_FMT_SP 0x200
+
+typedef struct {
+ int max_mq, min_mq, flag, min_baseQ, capQ_thres, max_depth;
+ char *reg, *fn_pos, *pl_list;
+ faidx_t *fai;
+ kh_64_t *hash;
+} mplp_conf_t;
+
+typedef struct {
+ bamFile fp;
+ bam_iter_t iter;
+ int min_mq, flag, ref_id, capQ_thres;
+ char *ref;
+} mplp_aux_t;
+
+typedef struct {
+ int n;
+ int *n_plp, *m_plp;
+ bam_pileup1_t **plp;
+} mplp_pileup_t;
+
+static int mplp_func(void *data, bam1_t *b)
+{
+ extern int bam_realn(bam1_t *b, const char *ref);
+ extern int bam_prob_realn_core(bam1_t *b, const char *ref, int);
+ extern int bam_cap_mapQ(bam1_t *b, char *ref, int thres);
+ mplp_aux_t *ma = (mplp_aux_t*)data;
+ int ret, skip = 0;
+ do {
+ int has_ref = (ma->ref && ma->ref_id == b->core.tid)? 1 : 0;
+ ret = ma->iter? bam_iter_read(ma->fp, ma->iter, b) : bam_read1(ma->fp, b);
+ if (ret < 0) break;
+ skip = 0;
+ if (has_ref && (ma->flag&MPLP_REALN)) bam_prob_realn_core(b, ma->ref, 1);
+ if (has_ref && ma->capQ_thres > 10) {
+ int q = bam_cap_mapQ(b, ma->ref, ma->capQ_thres);
+ if (q < 0) skip = 1;
+ else if (b->core.qual > q) b->core.qual = q;
+ } else if (b->core.flag&BAM_FUNMAP) skip = 1;
+ else if (b->core.qual < ma->min_mq) skip = 1;
+ else if ((ma->flag&MPLP_NO_ORPHAN) && (b->core.flag&1) && !(b->core.flag&2)) skip = 1;
+ } while (skip);
+ return ret;
+}
+
+static void group_smpl(mplp_pileup_t *m, bam_sample_t *sm, kstring_t *buf,
+ int n, char *const*fn, int *n_plp, const bam_pileup1_t **plp)
+{
+ int i, j;
+ memset(m->n_plp, 0, m->n * sizeof(int));
+ for (i = 0; i < n; ++i) {
+ for (j = 0; j < n_plp[i]; ++j) {
+ const bam_pileup1_t *p = plp[i] + j;
+ uint8_t *q;
+ int id = -1;
+ q = bam_aux_get(p->b, "RG");
+ if (q) id = bam_smpl_rg2smid(sm, fn[i], (char*)q+1, buf);
+ if (id < 0) id = bam_smpl_rg2smid(sm, fn[i], 0, buf);
+ assert(id >= 0 && id < m->n);
+ if (m->n_plp[id] == m->m_plp[id]) {
+ m->m_plp[id] = m->m_plp[id]? m->m_plp[id]<<1 : 8;
+ m->plp[id] = realloc(m->plp[id], sizeof(bam_pileup1_t) * m->m_plp[id]);
+ }
+ m->plp[id][m->n_plp[id]++] = *p;
+ }
+ }
+}
+
+static int mpileup(mplp_conf_t *conf, int n, char **fn)
+{
+ extern void *bcf_call_add_rg(void *rghash, const char *hdtext, const char *list);
+ extern void bcf_call_del_rghash(void *rghash);
+ mplp_aux_t **data;
+ int i, tid, pos, *n_plp, beg0 = 0, end0 = 1u<<29, ref_len, ref_tid, max_depth;
+ const bam_pileup1_t **plp;
+ bam_mplp_t iter;
+ bam_header_t *h = 0;
+ char *ref;
+ khash_t(64) *hash = 0;
+ void *rghash = 0;
+
+ bcf_callaux_t *bca = 0;
+ bcf_callret1_t *bcr = 0;
+ bcf_call_t bc;
+ bcf_t *bp = 0;
+ bcf_hdr_t *bh = 0;
+
+ bam_sample_t *sm = 0;
+ kstring_t buf;
+ mplp_pileup_t gplp;
+
+ memset(&gplp, 0, sizeof(mplp_pileup_t));
+ memset(&buf, 0, sizeof(kstring_t));
+ memset(&bc, 0, sizeof(bcf_call_t));
+ data = calloc(n, sizeof(void*));
+ plp = calloc(n, sizeof(void*));
+ n_plp = calloc(n, sizeof(int*));
+ sm = bam_smpl_init();
+
+ // read the header and initialize data
+ for (i = 0; i < n; ++i) {
+ bam_header_t *h_tmp;
+ data[i] = calloc(1, sizeof(mplp_aux_t));
+ data[i]->min_mq = conf->min_mq;
+ data[i]->flag = conf->flag;
+ data[i]->capQ_thres = conf->capQ_thres;
+ data[i]->fp = strcmp(fn[i], "-") == 0? bam_dopen(fileno(stdin), "r") : bam_open(fn[i], "r");
+ h_tmp = bam_header_read(data[i]->fp);
+ bam_smpl_add(sm, fn[i], h_tmp->text);
+ rghash = bcf_call_add_rg(rghash, h_tmp->text, conf->pl_list);
+ if (conf->reg) {
+ int beg, end;
+ bam_index_t *idx;
+ idx = bam_index_load(fn[i]);
+ if (idx == 0) {
+ fprintf(stderr, "[%s] fail to load index for %d-th input.\n", __func__, i+1);
+ exit(1);
+ }
+ if (bam_parse_region(h_tmp, conf->reg, &tid, &beg, &end) < 0) {
+ fprintf(stderr, "[%s] malformatted region or wrong seqname for %d-th input.\n", __func__, i+1);
+ exit(1);
+ }
+ if (i == 0) beg0 = beg, end0 = end;
+ data[i]->iter = bam_iter_query(idx, tid, beg, end);
+ bam_index_destroy(idx);
+ }
+ if (i == 0) h = h_tmp;
+ else {
+ // FIXME: to check consistency
+ bam_header_destroy(h_tmp);
+ }
+ }
+ gplp.n = sm->n;
+ gplp.n_plp = calloc(sm->n, sizeof(int));
+ gplp.m_plp = calloc(sm->n, sizeof(int));
+ gplp.plp = calloc(sm->n, sizeof(void*));
+
+ fprintf(stderr, "[%s] %d samples in %d input files\n", __func__, sm->n, n);
+ if (conf->fn_pos) hash = load_pos(conf->fn_pos, h);
+ // write the VCF header
+ if (conf->flag & MPLP_GLF) {
+ kstring_t s;
+ bh = calloc(1, sizeof(bcf_hdr_t));
+ s.l = s.m = 0; s.s = 0;
+ bp = bcf_open("-", (conf->flag&MPLP_NO_COMP)? "wu" : "w");
+ for (i = 0; i < h->n_targets; ++i) {
+ kputs(h->target_name[i], &s);
+ kputc('\0', &s);
+ }
+ bh->l_nm = s.l;
+ bh->name = malloc(s.l);
+ memcpy(bh->name, s.s, s.l);
+ s.l = 0;
+ for (i = 0; i < sm->n; ++i) {
+ kputs(sm->smpl[i], &s); kputc('\0', &s);
+ }
+ bh->l_smpl = s.l;
+ bh->sname = malloc(s.l);
+ memcpy(bh->sname, s.s, s.l);
+ bh->l_txt = 0;
+ free(s.s);
+ bcf_hdr_sync(bh);
+ bcf_hdr_write(bp, bh);
+ bca = bcf_call_init(-1., conf->min_baseQ);
+ bcr = calloc(sm->n, sizeof(bcf_callret1_t));
+ bca->rghash = rghash;
+ }
+ ref_tid = -1; ref = 0;
+ iter = bam_mplp_init(n, mplp_func, (void**)data);
+ max_depth = conf->max_depth;
+ if (max_depth * sm->n > 1<<20)
+ fprintf(stderr, "(%s) Max depth is above 1M. Potential memory hog!\n", __func__);
+ if (max_depth * sm->n < 8000) {
+ max_depth = 8000 / sm->n;
+ fprintf(stderr, "<%s> Set max per-sample depth to %d\n", __func__, max_depth);
+ }
+ bam_mplp_set_maxcnt(iter, max_depth);
+ while (bam_mplp_auto(iter, &tid, &pos, n_plp, plp) > 0) {
+ if (conf->reg && (pos < beg0 || pos >= end0)) continue; // out of the region requested
+ if (hash) {
+ khint_t k;
+ k = kh_get(64, hash, (uint64_t)tid<<32 | pos);
+ if (k == kh_end(hash)) continue;
+ }
+ if (tid != ref_tid) {
+ free(ref); ref = 0;
+ if (conf->fai) ref = fai_fetch(conf->fai, h->target_name[tid], &ref_len);
+ for (i = 0; i < n; ++i) data[i]->ref = ref, data[i]->ref_id = tid;
+ ref_tid = tid;
+ }
+ if (conf->flag & MPLP_GLF) {
+ int _ref0, ref16;
+ bcf1_t *b = calloc(1, sizeof(bcf1_t));
+ group_smpl(&gplp, sm, &buf, n, fn, n_plp, plp);
+ _ref0 = (ref && pos < ref_len)? ref[pos] : 'N';
+ ref16 = bam_nt16_table[_ref0];
+ for (i = 0; i < gplp.n; ++i)
+ bcf_call_glfgen(gplp.n_plp[i], gplp.plp[i], ref16, bca, bcr + i);
+ bcf_call_combine(gplp.n, bcr, ref16, &bc);
+ bcf_call2bcf(tid, pos, &bc, b, (conf->flag&(MPLP_FMT_DP|MPLP_FMT_SP))? bcr : 0,
+ (conf->flag&MPLP_FMT_SP), 0, 0);
+ bcf_write(bp, bh, b);
+ bcf_destroy(b);
+ // call indels
+ if (bcf_call_gap_prep(gplp.n, gplp.n_plp, gplp.plp, pos, bca, ref, rghash) >= 0) {
+ for (i = 0; i < gplp.n; ++i)
+ bcf_call_glfgen(gplp.n_plp[i], gplp.plp[i], -1, bca, bcr + i);
+ if (bcf_call_combine(gplp.n, bcr, -1, &bc) >= 0) {
+ b = calloc(1, sizeof(bcf1_t));
+ bcf_call2bcf(tid, pos, &bc, b, (conf->flag&(MPLP_FMT_DP|MPLP_FMT_SP))? bcr : 0,
+ (conf->flag&MPLP_FMT_SP), bca, ref);
+ bcf_write(bp, bh, b);
+ bcf_destroy(b);
+ }
+ }
+ } else {
+ printf("%s\t%d\t%c", h->target_name[tid], pos + 1, (ref && pos < ref_len)? ref[pos] : 'N');
+ for (i = 0; i < n; ++i) {
+ int j;
+ printf("\t%d\t", n_plp[i]);
+ if (n_plp[i] == 0) printf("*\t*");
+ else {
+ for (j = 0; j < n_plp[i]; ++j)
+ pileup_seq(plp[i] + j, pos, ref_len, ref);
+ putchar('\t');
+ for (j = 0; j < n_plp[i]; ++j) {
+ const bam_pileup1_t *p = plp[i] + j;
+ int c = bam1_qual(p->b)[p->qpos] + 33;
+ if (c > 126) c = 126;
+ putchar(c);
+ }
+ }
+ }
+ putchar('\n');
+ }
+ }
+
+ bcf_close(bp);
+ bam_smpl_destroy(sm); free(buf.s);
+ for (i = 0; i < gplp.n; ++i) free(gplp.plp[i]);
+ free(gplp.plp); free(gplp.n_plp); free(gplp.m_plp);
+ bcf_call_del_rghash(rghash);
+ if (hash) { // free the hash table
+ khint_t k;
+ for (k = kh_begin(hash); k < kh_end(hash); ++k)
+ if (kh_exist(hash, k)) free(kh_val(hash, k));
+ kh_destroy(64, hash);
+ }
+ bcf_hdr_destroy(bh); bcf_call_destroy(bca); free(bc.PL); free(bcr);
+ bam_mplp_destroy(iter);
+ bam_header_destroy(h);
+ for (i = 0; i < n; ++i) {
+ bam_close(data[i]->fp);
+ if (data[i]->iter) bam_iter_destroy(data[i]->iter);
+ free(data[i]);
+ }
+ free(data); free(plp); free(ref); free(n_plp);
+ return 0;
+}
+
+#define MAX_PATH_LEN 1024
+int read_file_list(const char *file_list,int *n,char **argv[])
+{
+ char buf[MAX_PATH_LEN];
+ int len, nfiles;
+ char **files;
+
+ FILE *fh = fopen(file_list,"r");
+ if ( !fh )
+ {
+ fprintf(stderr,"%s: %s\n", file_list,strerror(errno));
+ return 1;
+ }
+
+ // Speed is not an issue here, determine the number of files by reading the file twice
+ nfiles = 0;
+ while ( fgets(buf,MAX_PATH_LEN,fh) ) nfiles++;
+
+ if ( fseek(fh, 0L, SEEK_SET) )
+ {
+ fprintf(stderr,"%s: %s\n", file_list,strerror(errno));
+ return 1;
+ }
+
+ files = calloc(nfiles,sizeof(char*));
+ nfiles = 0;
+ while ( fgets(buf,MAX_PATH_LEN,fh) )
+ {
+ len = strlen(buf);
+ while ( len>0 && isspace(buf[len-1]) ) len--;
+ if ( !len ) continue;
+
+ files[nfiles] = malloc(sizeof(char)*(len+1));
+ strncpy(files[nfiles],buf,len);
+ files[nfiles][len] = 0;
+ nfiles++;
+ }
+ fclose(fh);
+ if ( !nfiles )
+ {
+ fprintf(stderr,"No files read from %s\n", file_list);
+ return 1;
+ }
+ *argv = files;
+ *n = nfiles;
+ return 0;
+}
+#undef MAX_PATH_LEN
+
+int bam_mpileup(int argc, char *argv[])
+{
+ int c;
+ const char *file_list = NULL;
+ char **fn = NULL;
+ int nfiles = 0;
+ mplp_conf_t mplp;
+ memset(&mplp, 0, sizeof(mplp_conf_t));
+ mplp.max_mq = 60;
+ mplp.min_baseQ = 13;
+ mplp.capQ_thres = 0;
+ mplp.max_depth = 250;
+ mplp.flag = MPLP_NO_ORPHAN | MPLP_REALN;
+ while ((c = getopt(argc, argv, "gf:r:l:M:q:Q:uaORC:BDSd:b:P:")) >= 0) {
+ switch (c) {
+ case 'f':
+ mplp.fai = fai_load(optarg);
+ if (mplp.fai == 0) return 1;
+ break;
+ case 'd': mplp.max_depth = atoi(optarg); break;
+ case 'r': mplp.reg = strdup(optarg); break;
+ case 'l': mplp.fn_pos = strdup(optarg); break;
+ case 'P': mplp.pl_list = strdup(optarg); break;
+ case 'g': mplp.flag |= MPLP_GLF; break;
+ case 'u': mplp.flag |= MPLP_NO_COMP | MPLP_GLF; break;
+ case 'a': mplp.flag |= MPLP_NO_ORPHAN | MPLP_REALN; break;
+ case 'B': mplp.flag &= ~MPLP_REALN & ~MPLP_NO_ORPHAN; break;
+ case 'O': mplp.flag |= MPLP_NO_ORPHAN; break;
+ case 'R': mplp.flag |= MPLP_REALN; break;
+ case 'D': mplp.flag |= MPLP_FMT_DP; break;
+ case 'S': mplp.flag |= MPLP_FMT_SP; break;
+ case 'C': mplp.capQ_thres = atoi(optarg); break;
+ case 'M': mplp.max_mq = atoi(optarg); break;
+ case 'q': mplp.min_mq = atoi(optarg); break;
+ case 'Q': mplp.min_baseQ = atoi(optarg); break;
+ case 'b': file_list = optarg; break;
+ }
+ }
+ if (argc == 1) {
+ fprintf(stderr, "\n");
+ fprintf(stderr, "Usage: samtools mpileup [options] in1.bam [in2.bam [...]]\n\n");
+ fprintf(stderr, "Options: -f FILE reference sequence file [null]\n");
+ fprintf(stderr, " -r STR region in which pileup is generated [null]\n");
+ fprintf(stderr, " -l FILE list of positions (format: chr pos) [null]\n");
+ fprintf(stderr, " -b FILE list of input BAM files [null]\n");
+ fprintf(stderr, " -M INT cap mapping quality at INT [%d]\n", mplp.max_mq);
+ fprintf(stderr, " -Q INT min base quality [%d]\n", mplp.min_baseQ);
+ fprintf(stderr, " -q INT filter out alignment with MQ smaller than INT [%d]\n", mplp.min_mq);
+ fprintf(stderr, " -d INT max per-sample depth [%d]\n", mplp.max_depth);
+ fprintf(stderr, " -P STR comma separated list of platforms for indels [all]\n");
+ fprintf(stderr, " -g generate BCF output\n");
+ fprintf(stderr, " -u do not compress BCF output\n");
+ fprintf(stderr, " -B disable BAQ computation\n");
+ fprintf(stderr, " -D output per-sample DP\n");
+ fprintf(stderr, " -S output per-sample SP (strand bias P-value, slow)\n");
+ fprintf(stderr, "\n");
+ fprintf(stderr, "Notes: Assuming diploid individuals.\n\n");
+ return 1;
+ }
+ if ( file_list )
+ {
+ if ( read_file_list(file_list,&nfiles,&fn) ) return 1;
+ mpileup(&mplp,nfiles,fn);
+ for (c=0; c<nfiles; c++) free(fn[c]);
+ free(fn);
+ }
+ else
+ mpileup(&mplp, argc - optind, argv + optind);
+ free(mplp.reg); free(mplp.pl_list);
+ if (mplp.fai) fai_destroy(mplp.fai);
+ return 0;
+}