#define MINUS_CONST 0x10000000
#define INDEL_WINDOW_SIZE 50
-#define MAX_SCORE 90
+#define MIN_SUPPORT_COEF 500
void *bcf_call_add_rg(void *_hash, const char *hdtext, const char *list)
{
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, n_alt;
+ int i, s, j, k, t, n_types, *types, max_rd_len, left, right, max_ins, *score1, *score2;
+ int 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;
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;
+ int m, n_alt = 0, n_tot = 0;
uint32_t *aux;
aux = calloc(N + 1, 4);
m = max_rd_len = 0;
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 && (rghash == 0 || p->aux == 0))
- aux[m++] = MINUS_CONST + p->indel;
+ if (rghash == 0 || p->aux == 0) {
+ ++n_tot;
+ if (p->indel != 0) {
+ ++n_alt;
+ 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;
}
// squeeze out identical types
for (i = 1, n_types = 1; i < m; ++i)
if (aux[i] != aux[i-1]) ++n_types;
- if (n_types == 1) { // no indels
+ if (n_types == 1 || n_alt * MIN_SUPPORT_COEF < n_tot) { // no indels or too few supporting reads
free(aux); return -1;
}
types = (int*)calloc(n_types, sizeof(int));
// compute the likelihood given each type of indel for each read
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));
+ score1 = calloc(N * n_types, sizeof(int));
+ score2 = calloc(N * n_types, sizeof(int));
bca->indelreg = 0;
for (t = 0; t < n_types; ++t) {
int l, ir;
- ka_param2_t ap = ka_param2_qual;
- ap.band_width = abs(types[t]) + 3;
+ kpa_par_t apf1 = { 1e-4, 1e-2, 10 }, apf2 = { 1e-6, 1e-3, 10 };
+ apf1.bw = apf2.bw = 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]);
bam_pileup1_t *p = plp[s] + i;
int qbeg, qend, tbeg, tend, sc;
uint8_t *seq = bam1_seq(p->b);
+ // FIXME: the following skips soft clips, but using them may be more sensitive.
// 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);
// 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
- if (0) {
- uint8_t *qq = calloc(qend - qbeg, 1);
- for (l = 0; l < qend - qbeg; ++l) qq[l] = 23;
+ { // do realignment; this is the bottleneck
+ const uint8_t *qual = bam1_qual(p->b), *bq;
+ uint8_t *qq;
+ qq = calloc(qend - qbeg, 1);
+ bq = (uint8_t*)bam_aux_get(p->b, "ZQ");
+ if (bq) ++bq; // skip type
+ for (l = qbeg; l < qend; ++l) {
+ qq[l - qbeg] = bq? qual[l] + (bq[l] - 64) : qual[l];
+ if (qq[l - qbeg] > 30) qq[l - qbeg] = 30;
+ if (qq[l - qbeg] < 7) qq[l - qbeg] = 7;
+ }
sc = kpa_glocal((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
- (uint8_t*)query, qend - qbeg, qq, &kpa_par_alt, 0, 0);
- score[K*n_types + t] = sc;
- } else {
- sc = ka_global_score((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
- (uint8_t*)query, qend - qbeg, &ap);
- score[K*n_types + t] = -sc;
+ (uint8_t*)query, qend - qbeg, qq, &apf1, 0, 0);
+ l = (int)(100. * sc / (qend - qbeg) + .499); // used for adjusting indelQ below
+ if (l > 255) l = 255;
+ score1[K*n_types + t] = score2[K*n_types + t] = sc<<8 | l;
+ if (sc > 5) {
+ sc = kpa_glocal((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
+ (uint8_t*)query, qend - qbeg, qq, &apf2, 0, 0);
+ l = (int)(100. * sc / (qend - qbeg) + .499);
+ if (l > 255) l = 255;
+ score2[K*n_types + t] = sc<<8 | l;
+ }
+ free(qq);
}
/*
for (l = 0; l < tend - tbeg + abs(types[t]); ++l)
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, seqQ;
+ int *sct = &score1[K*n_types], indelQ1, indelQ2, seqQ, indelQ;
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)
* compromise for multi-allelic indels.
*/
if ((sc[0]&0x3f) == ref_type) {
- indelQ = (sc[1]>>6) - (sc[0]>>6);
+ indelQ1 = (sc[1]>>14) - (sc[0]>>14);
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);
+ indelQ1 = (sc[t]>>14) - (sc[0]>>14);
seqQ = est_seqQ(bca, types[sc[0]&0x3f], l_run);
}
- if (sc[0]>>6 > MAX_SCORE) indelQ = 0; // too many mismatches; something bad possibly happened
+ tmp = sc[0]>>6 & 0xff;
+ indelQ1 = tmp > 111? 0 : (int)((1. - tmp/111.) * indelQ1 + .499); // reduce indelQ
+ sct = &score2[K*n_types];
+ 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)
+ tmp = sc[j], sc[j] = sc[j-1], sc[j-1] = tmp;
+ if ((sc[0]&0x3f) == ref_type) {
+ indelQ2 = (sc[1]>>14) - (sc[0]>>14);
+ } else {
+ for (t = 0; t < n_types; ++t) // look for the reference type
+ if ((sc[t]&0x3f) == ref_type) break;
+ indelQ2 = (sc[t]>>14) - (sc[0]>>14);
+ }
+ tmp = sc[0]>>6 & 0xff;
+ indelQ2 = tmp > 111? 0 : (int)((1. - tmp/111.) * indelQ2 + .499);
+ // pick the smaller between indelQ1 and indelQ2
+ indelQ = indelQ1 < indelQ2? indelQ1 : indelQ2;
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);
}
}
}
- free(score);
+ free(score1); free(score2);
// free
free(types); free(inscns);
return n_alt > 0? 0 : -1;