bcf1_t *vcfbuf_max_ld(vcfbuf_t *buf, bcf1_t *rec, double *ld)
{
*ld = -1;
if ( !buf->rbuf.n ) return NULL;
int i = buf->rbuf.f;
// Relying on vcfbuf being properly flushed - all sites in the buffer
// must come from the same chromosome
if ( buf->vcf[i].rec->rid != rec->rid ) return NULL;
int imax = 0;
double max = 0;
for (i=-1; rbuf_next(&buf->rbuf,&i); )
{
if ( buf->ld.skip_filter )
{
if ( buf->vcf[i].rec->d.n_flt > 1 ) continue; // multiple filters are set
if ( buf->vcf[i].rec->d.n_flt==1 && buf->vcf[i].rec->d.flt[0]!=0 ) continue; // not PASS
}
double val = _calc_ld(buf, buf->vcf[i].rec, rec);
if ( buf->ld.max && buf->ld.max < val )
{
*ld = val;
return buf->vcf[i].rec;
}
if ( val > max )
{
max = val;
imax = i;
}
}
*ld = max;
return buf->vcf[imax].rec;
}
static void _prune_sites(vcfbuf_t *buf, int flush_all)
{
int nbuf = flush_all ? buf->rbuf.n : buf->rbuf.n - 1;
if ( nbuf > buf->prune.mvrec )
{
buf->prune.idx = (int*) realloc(buf->prune.idx, nbuf*sizeof(int));
buf->prune.vrec = (vcfrec_t**) realloc(buf->prune.vrec, nbuf*sizeof(vcfrec_t*));
buf->prune.mvrec = nbuf;
}
// set allele frequency and prepare buffer for sorting
int i,k,irec = 0;
for (i=-1; rbuf_next(&buf->rbuf,&i) && irec<nbuf; )
{
bcf1_t *line = buf->vcf[i].rec;
if ( line->n_allele > buf->prune.mac )
{
buf->prune.ac = (int*) realloc(buf->prune.ac, line->n_allele*sizeof(*buf->prune.ac));
buf->prune.mac = line->n_allele;
}
if ( !buf->vcf[i].af_set )
{
buf->vcf[i].af = 0;
if ( buf->prune.af_tag )
{
if ( bcf_get_info_float(buf->hdr,line,buf->prune.af_tag,&buf->prune.farr, &buf->prune.mfarr) > 0 ) buf->vcf[i].af = buf->prune.farr[0];
}
else if ( bcf_calc_ac(buf->hdr, line, buf->prune.ac, BCF_UN_INFO|BCF_UN_FMT) )
{
int ntot = buf->prune.ac[0], nalt = 0;
for (k=1; k<line->n_allele; k++) nalt += buf->prune.ac[k];
buf->vcf[i].af = ntot ? (float)nalt/ntot : 0;
}
buf->vcf[i].af_set = 1;
}
buf->vcf[i].idx = irec;
buf->prune.vrec[irec++] = &buf->vcf[i];
}
// sort by allele frequency, low AF will be removed preferentially
qsort(buf->prune.vrec, nbuf, sizeof(*buf->prune.vrec), cmpvrec);
// sort the rbuf indexes to be pruned descendently so that j-th rbuf index
// is removed before i-th index if i<j
int nprune = nbuf - buf->prune.max_sites;
for (i=0; i<nprune; i++)
buf->prune.idx[i] = buf->prune.vrec[i]->idx;
qsort(buf->prune.idx, nprune, sizeof(int), cmpint_desc);
for (i=0; i<nprune; i++)
rbuf_remove_kth(&buf->rbuf, vcfrec_t, buf->prune.idx[i], buf->vcf);
}
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);
}
}
