static int _overlap_can_flush(vcfbuf_t *buf, int flush_all) { if ( flush_all ) { buf->overlap.rid = -1; return 1; } int i = rbuf_last(&buf->rbuf); vcfrec_t *last = &buf->vcf[i]; if ( buf->overlap.rid != last->rec->rid ) buf->overlap.end = 0; int beg_pos = last->rec->pos; int end_pos = last->rec->pos + last->rec->rlen - 1; // Assuming left-aligned indels. In case it is a deletion, the real variant // starts one base after. If an insertion, the overlap with previous zero length. int imin = last->rec->rlen; for (i=0; i<last->rec->n_allele; i++) { char *ref = last->rec->d.allele[0]; char *alt = last->rec->d.allele[i]; if ( *alt == '<' ) continue; // ignore symbolic alleles while ( *ref && *alt && nt_to_upper(*ref)==nt_to_upper(*alt) ) { ref++; alt++; } if ( imin > ref - last->rec->d.allele[0] ) imin = ref - last->rec->d.allele[0]; } if ( beg_pos <= buf->overlap.end ) { beg_pos += imin; if ( beg_pos > end_pos ) end_pos = beg_pos; } if ( buf->rbuf.n==1 ) { buf->overlap.rid = last->rec->rid; buf->overlap.end = end_pos; return 0; } if ( beg_pos <= buf->overlap.end ) { if ( buf->overlap.end < end_pos ) buf->overlap.end = end_pos; return 0; } return 1; }
bcf1_t *vcfbuf_flush(vcfbuf_t *buf, int flush_all) { int i,j; if ( buf->rbuf.n==0 ) return NULL; if ( flush_all ) goto ret; i = rbuf_kth(&buf->rbuf, 0); // first j = rbuf_last(&buf->rbuf); // last if ( buf->vcf[i].rec->rid != buf->vcf[j].rec->rid ) goto ret; if ( buf->overlap.active ) { int ret = _overlap_can_flush(buf, flush_all); //printf("can_flush: %d %d - %d\n", ret, buf->vcf[i].rec->pos+1, buf->vcf[j].rec->pos+1); if ( ret ) goto ret; } //if ( buf->overlap.active && _overlap_can_flush(buf, flush_all) ) goto ret; if ( buf->win > 0 ) { if ( buf->rbuf.n <= buf->win ) return NULL; goto ret; } else if ( buf->win < 0 ) { if ( buf->vcf[i].rec->pos - buf->vcf[j].rec->pos > buf->win ) return NULL; } else return NULL; ret: if ( buf->prune.max_sites && buf->prune.max_sites < buf->rbuf.n ) _prune_sites(buf, flush_all); i = rbuf_shift(&buf->rbuf); return buf->vcf[i].rec; }
static void buffered_filters(args_t *args, bcf1_t *line) { /** * The logic of SnpGap=3. The SNPs at positions 1 and 7 are filtered, * positions 0 and 8 are not: * 0123456789 * ref .G.GT..G.. * del .A.G-..A.. * Here the positions 1 and 6 are filtered, 0 and 7 are not: * 0123-456789 * ref .G.G-..G.. * ins .A.GT..A.. * * The logic of IndelGap=2. The second indel is filtered: * 012345678901 * ref .GT.GT..GT.. * del .G-.G-..G-.. * And similarly here, the second is filtered: * 01 23 456 78 * ref .A-.A-..A-.. * ins .AT.AT..AT.. */ // To avoid additional data structure, we abuse bcf1_t's var and var_type records. const int SnpGap_set = VCF_OTHER<<1; const int IndelGap_set = VCF_OTHER<<2; const int IndelGap_flush = VCF_OTHER<<3; int var_type = 0, i; if ( line ) { // Still on the same chromosome? int ilast = rbuf_last(&args->rbuf); if ( ilast>=0 && line->rid != args->rbuf_lines[ilast]->rid ) flush_buffer(args, args->rbuf.n); // new chromosome, flush everything rbuf_expand0(&args->rbuf,bcf1_t*,args->rbuf.n,args->rbuf_lines); // Insert the new record in the buffer. The line would be overwritten in // the next bcf_sr_next_line call, therefore we need to swap it with an // unused one ilast = rbuf_append(&args->rbuf); if ( !args->rbuf_lines[ilast] ) args->rbuf_lines[ilast] = bcf_init1(); SWAP(bcf1_t*, args->files->readers[0].buffer[0], args->rbuf_lines[ilast]); var_type = bcf_get_variant_types(line); // Find out the size of an indel. The indel boundaries are based on REF // (POS+1,POS+rlen-1). This is not entirely correct: mpileup likes to // output REF=CAGAGAGAGA, ALT=CAGAGAGAGAGA where REF=C,ALT=CGA could be // used. This filter is therefore more strict and may remove some valid // SNPs. int len = 1; if ( var_type & VCF_INDEL ) { for (i=1; i<line->n_allele; i++) if ( len < 1-line->d.var[i].n ) len = 1-line->d.var[i].n; } // Set the REF allele's length to max deletion length or to 1 if a SNP or an insertion. line->d.var[0].n = len; } int k_flush = 1; if ( args->indel_gap ) { k_flush = 0; // Find indels which are too close to each other int last_to = -1; for (i=-1; rbuf_next(&args->rbuf,&i); ) { bcf1_t *rec = args->rbuf_lines[i]; int rec_from = rec->pos; if ( last_to!=-1 && last_to < rec_from ) break; k_flush++; if ( !(rec->d.var_type & VCF_INDEL) ) continue; rec->d.var_type |= IndelGap_set; last_to = args->indel_gap + rec->pos + rec->d.var[0].n - 1; } if ( i==args->rbuf.f && line && last_to!=-1 ) k_flush = 0; if ( k_flush || !line ) { // Select the best indel from the cluster of k_flush indels int k = 0, max_ac = -1, imax_ac = -1; for (i=-1; rbuf_next(&args->rbuf,&i) && k<k_flush; ) { k++; bcf1_t *rec = args->rbuf_lines[i]; if ( !(rec->d.var_type & IndelGap_set) ) continue; hts_expand(int, rec->n_allele, args->ntmpi, args->tmpi); int ret = bcf_calc_ac(args->hdr, rec, args->tmpi, BCF_UN_ALL); if ( imax_ac==-1 || (ret && max_ac < args->tmpi[1]) ) { max_ac = args->tmpi[1]; imax_ac = i; } } // Filter all but the best indel (with max AF or first if AF not available) k = 0; for (i=-1; rbuf_next(&args->rbuf,&i) && k<k_flush; ) { k++; bcf1_t *rec = args->rbuf_lines[i]; if ( !(rec->d.var_type & IndelGap_set) ) continue; rec->d.var_type |= IndelGap_flush; if ( i!=imax_ac ) bcf_add_filter(args->hdr, args->rbuf_lines[i], args->IndelGap_id); } }