10 KHASH_SET_INIT_STR(rg)
12 #define MINUS_CONST 0x10000000
13 #define INDEL_WINDOW_SIZE 50
14 #define MIN_SUPPORT_COEF 500
16 void *bcf_call_add_rg(void *_hash, const char *hdtext, const char *list)
18 const char *s, *p, *q, *r, *t;
20 if (list == 0 || hdtext == 0) return _hash;
21 if (_hash == 0) _hash = kh_init(rg);
22 hash = (khash_t(rg)*)_hash;
23 if ((s = strstr(hdtext, "@RG\t")) == 0) return hash;
25 t = strstr(s + 4, "@RG\t"); // the next @RG
26 if ((p = strstr(s, "\tID:")) != 0) p += 4;
27 if ((q = strstr(s, "\tPL:")) != 0) q += 4;
28 if (p && q && (t == 0 || (p < t && q < t))) { // ID and PL are both present
31 for (r = p; *r && *r != '\t' && *r != '\n'; ++r); lp = r - p;
32 for (r = q; *r && *r != '\t' && *r != '\n'; ++r); lq = r - q;
33 x = calloc((lp > lq? lp : lq) + 1, 1);
34 for (r = q; *r && *r != '\t' && *r != '\n'; ++r) x[r-q] = *r;
35 if (strstr(list, x)) { // insert ID to the hash table
38 for (r = p; *r && *r != '\t' && *r != '\n'; ++r) x[r-p] = *r;
40 k = kh_get(rg, hash, x);
41 if (k == kh_end(hash)) k = kh_put(rg, hash, x, &ret);
50 void bcf_call_del_rghash(void *_hash)
53 khash_t(rg) *hash = (khash_t(rg)*)_hash;
54 if (hash == 0) return;
55 for (k = kh_begin(hash); k < kh_end(hash); ++k)
56 if (kh_exist(hash, k))
57 free((char*)kh_key(hash, k));
61 static int tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, int is_left, int32_t *_tpos)
63 int k, x = c->pos, y = 0, last_y = 0;
65 for (k = 0; k < c->n_cigar; ++k) {
66 int op = cigar[k] & BAM_CIGAR_MASK;
67 int l = cigar[k] >> BAM_CIGAR_SHIFT;
68 if (op == BAM_CMATCH) {
69 if (c->pos > tpos) return y;
72 return y + (tpos - x);
76 } else if (op == BAM_CINS || op == BAM_CSOFT_CLIP) y += l;
77 else if (op == BAM_CDEL || op == BAM_CREF_SKIP) {
79 *_tpos = is_left? x : x + l;
88 // FIXME: check if the inserted sequence is consistent with the homopolymer run
89 // l is the relative gap length and l_run is the length of the homopolymer on the reference
90 static inline int est_seqQ(const bcf_callaux_t *bca, int l, int l_run)
93 q = bca->openQ + bca->extQ * (abs(l) - 1);
94 qh = l_run >= 3? (int)(bca->tandemQ * (double)abs(l) / l_run + .499) : 1000;
95 return q < qh? q : qh;
98 static inline int est_indelreg(int pos, const char *ref, int l, char *ins4)
100 int i, j, max = 0, max_i = pos, score = 0;
102 for (i = pos + 1, j = 0; ref[i]; ++i, ++j) {
103 if (ins4) score += (toupper(ref[i]) != "ACGTN"[(int)ins4[j%l]])? -10 : 1;
104 else score += (toupper(ref[i]) != toupper(ref[pos+1+j%l]))? -10 : 1;
105 if (score < 0) break;
106 if (max < score) max = score, max_i = i;
111 int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_callaux_t *bca, const char *ref,
114 extern void ks_introsort_uint32_t(int, uint32_t*);
115 int i, s, j, k, t, n_types, *types, max_rd_len, left, right, max_ins, *score1, *score2;
116 int N, K, l_run, ref_type, n_alt;
117 char *inscns = 0, *ref2, *query;
118 khash_t(rg) *hash = (khash_t(rg)*)rghash;
119 if (ref == 0 || bca == 0) return -1;
120 // mark filtered reads
123 for (s = N = 0; s < n; ++s) {
124 for (i = 0; i < n_plp[s]; ++i) {
125 bam_pileup1_t *p = plp[s] + i;
126 const uint8_t *rg = bam_aux_get(p->b, "RG");
127 p->aux = 1; // filtered by default
129 khint_t k = kh_get(rg, hash, (const char*)(rg + 1));
130 if (k != kh_end(hash)) p->aux = 0, ++N; // not filtered
134 if (N == 0) return -1; // no reads left
136 // determine if there is a gap
137 for (s = N = 0; s < n; ++s) {
138 for (i = 0; i < n_plp[s]; ++i)
139 if (plp[s][i].indel != 0) break;
140 if (i < n_plp[s]) break;
142 if (s == n) return -1; // there is no indel at this position.
143 for (s = N = 0; s < n; ++s) N += n_plp[s]; // N is the total number of reads
144 { // find out how many types of indels are present
145 int m, n_alt = 0, n_tot = 0;
147 aux = calloc(N + 1, 4);
149 aux[m++] = MINUS_CONST; // zero indel is always a type
150 for (s = 0; s < n; ++s) {
151 for (i = 0; i < n_plp[s]; ++i) {
152 const bam_pileup1_t *p = plp[s] + i;
153 if (rghash == 0 || p->aux == 0) {
157 aux[m++] = MINUS_CONST + p->indel;
160 j = bam_cigar2qlen(&p->b->core, bam1_cigar(p->b));
161 if (j > max_rd_len) max_rd_len = j;
164 ks_introsort(uint32_t, m, aux);
165 // squeeze out identical types
166 for (i = 1, n_types = 1; i < m; ++i)
167 if (aux[i] != aux[i-1]) ++n_types;
168 if (n_types == 1 || n_alt * MIN_SUPPORT_COEF < n_tot) { // no indels or too few supporting reads
169 free(aux); return -1;
171 types = (int*)calloc(n_types, sizeof(int));
173 types[t++] = aux[0] - MINUS_CONST;
174 for (i = 1; i < m; ++i)
175 if (aux[i] != aux[i-1])
176 types[t++] = aux[i] - MINUS_CONST;
178 for (t = 0; t < n_types; ++t)
179 if (types[t] == 0) break;
180 ref_type = t; // the index of the reference type (0)
181 assert(n_types < 64);
183 { // calculate left and right boundary
184 left = pos > INDEL_WINDOW_SIZE? pos - INDEL_WINDOW_SIZE : 0;
185 right = pos + INDEL_WINDOW_SIZE;
186 if (types[0] < 0) right -= types[0];
187 // in case the alignments stand out the reference
188 for (i = pos; i < right; ++i)
189 if (ref[i] == 0) break;
192 { // the length of the homopolymer run around the current position
193 int c = bam_nt16_table[(int)ref[pos + 1]];
194 if (c == 15) l_run = 1;
196 for (i = pos + 2; ref[i]; ++i)
197 if (bam_nt16_table[(int)ref[i]] != c) break;
199 for (i = pos; i >= 0; --i)
200 if (bam_nt16_table[(int)ref[i]] != c) break;
204 // construct the consensus sequence
205 max_ins = types[n_types - 1]; // max_ins is at least 0
207 int *inscns_aux = calloc(4 * n_types * max_ins, sizeof(int));
208 // count the number of occurrences of each base at each position for each type of insertion
209 for (t = 0; t < n_types; ++t) {
211 for (s = 0; s < n; ++s) {
212 for (i = 0; i < n_plp[s]; ++i) {
213 bam_pileup1_t *p = plp[s] + i;
214 if (p->indel == types[t]) {
215 uint8_t *seq = bam1_seq(p->b);
216 for (k = 1; k <= p->indel; ++k) {
217 int c = bam_nt16_nt4_table[bam1_seqi(seq, p->qpos + k)];
218 if (c < 4) ++inscns_aux[(t*max_ins+(k-1))*4 + c];
225 // use the majority rule to construct the consensus
226 inscns = calloc(n_types * max_ins, 1);
227 for (t = 0; t < n_types; ++t) {
228 for (j = 0; j < types[t]; ++j) {
229 int max = 0, max_k = -1, *ia = &inscns_aux[(t*max_ins+j)*4];
230 for (k = 0; k < 4; ++k)
232 max = ia[k], max_k = k;
233 inscns[t*max_ins + j] = max? max_k : 4;
238 // compute the likelihood given each type of indel for each read
239 ref2 = calloc(right - left + max_ins + 2, 1);
240 query = calloc(right - left + max_rd_len + max_ins + 2, 1);
241 score1 = calloc(N * n_types, sizeof(int));
242 score2 = calloc(N * n_types, sizeof(int));
244 for (t = 0; t < n_types; ++t) {
246 kpa_par_t apf1 = { 1e-4, 1e-2, 10 }, apf2 = { 1e-6, 1e-3, 10 };
247 apf1.bw = apf2.bw = abs(types[t]) + 3;
249 if (types[t] == 0) ir = 0;
250 else if (types[t] > 0) ir = est_indelreg(pos, ref, types[t], &inscns[t*max_ins]);
251 else ir = est_indelreg(pos, ref, -types[t], 0);
252 if (ir > bca->indelreg) bca->indelreg = ir;
253 // fprintf(stderr, "%d, %d, %d\n", pos, types[t], ir);
255 for (k = 0, j = left; j <= pos; ++j)
256 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
257 if (types[t] <= 0) j += -types[t];
258 else for (l = 0; l < types[t]; ++l)
259 ref2[k++] = inscns[t*max_ins + l];
260 if (types[0] < 0) { // mask deleted sequences to avoid a particular error in the model.
261 int jj, tmp = types[t] >= 0? -types[0] : -types[0] + types[t];
262 for (jj = 0; jj < tmp && j < right && ref[j]; ++jj, ++j)
265 for (; j < right && ref[j]; ++j)
266 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
267 if (j < right) right = j;
268 // align each read to ref2
269 for (s = K = 0; s < n; ++s) {
270 for (i = 0; i < n_plp[s]; ++i, ++K) {
271 bam_pileup1_t *p = plp[s] + i;
272 int qbeg, qend, tbeg, tend, sc;
273 uint8_t *seq = bam1_seq(p->b);
274 // FIXME: the following skips soft clips, but using them may be more sensitive.
275 // determine the start and end of sequences for alignment
276 qbeg = tpos2qpos(&p->b->core, bam1_cigar(p->b), left, 0, &tbeg);
277 qend = tpos2qpos(&p->b->core, bam1_cigar(p->b), right, 1, &tend);
280 tbeg = tbeg - l > left? tbeg - l : left;
282 // write the query sequence
283 for (l = qbeg; l < qend; ++l)
284 query[l - qbeg] = bam_nt16_nt4_table[bam1_seqi(seq, l)];
285 { // do realignment; this is the bottleneck
286 const uint8_t *qual = bam1_qual(p->b), *bq;
288 qq = calloc(qend - qbeg, 1);
289 bq = (uint8_t*)bam_aux_get(p->b, "BQ");
290 if (bq) ++bq; // skip type
291 for (l = qbeg; l < qend; ++l) {
292 qq[l - qbeg] = bq? qual[l] + (bq[l] - 33) : qual[l];
293 if (qq[l - qbeg] > 30) qq[l - qbeg] = 30;
294 if (qq[l - qbeg] < 7) qq[l - qbeg] = 7;
296 sc = kpa_glocal((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
297 (uint8_t*)query, qend - qbeg, qq, &apf1, 0, 0);
298 l = (int)(100. * sc / (qend - qbeg) + .499); // used for adjusting indelQ below
299 if (l > 255) l = 255;
300 score1[K*n_types + t] = score2[K*n_types + t] = sc<<8 | l;
302 sc = kpa_glocal((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
303 (uint8_t*)query, qend - qbeg, qq, &apf2, 0, 0);
304 l = (int)(100. * sc / (qend - qbeg) + .499);
305 if (l > 255) l = 255;
306 score2[K*n_types + t] = sc<<8 | l;
311 for (l = 0; l < tend - tbeg + abs(types[t]); ++l)
312 fputc("ACGTN"[(int)ref2[tbeg-left+l]], stderr);
314 for (l = 0; l < qend - qbeg; ++l) fputc("ACGTN"[(int)query[l]], stderr);
316 fprintf(stderr, "pos=%d type=%d read=%d:%d name=%s qbeg=%d tbeg=%d score=%d\n", pos, types[t], s, i, bam1_qname(p->b), qbeg, tbeg, sc);
321 free(ref2); free(query);
324 sc = alloca(n_types * sizeof(int));
325 sumq = alloca(n_types * sizeof(int));
326 memset(sumq, 0, sizeof(int) * n_types);
327 for (s = K = 0; s < n; ++s) {
328 for (i = 0; i < n_plp[s]; ++i, ++K) {
329 bam_pileup1_t *p = plp[s] + i;
330 int *sct = &score1[K*n_types], indelQ1, indelQ2, seqQ, indelQ;
331 for (t = 0; t < n_types; ++t) sc[t] = sct[t]<<6 | t;
332 for (t = 1; t < n_types; ++t) // insertion sort
333 for (j = t; j > 0 && sc[j] < sc[j-1]; --j)
334 tmp = sc[j], sc[j] = sc[j-1], sc[j-1] = tmp;
335 /* errmod_cal() assumes that if the call is wrong, the
336 * likelihoods of other events are equal. This is about
337 * right for substitutions, but is not desired for
338 * indels. To reuse errmod_cal(), I have to make
339 * compromise for multi-allelic indels.
341 if ((sc[0]&0x3f) == ref_type) {
342 indelQ1 = (sc[1]>>14) - (sc[0]>>14);
343 seqQ = est_seqQ(bca, types[sc[1]&0x3f], l_run);
345 for (t = 0; t < n_types; ++t) // look for the reference type
346 if ((sc[t]&0x3f) == ref_type) break;
347 indelQ1 = (sc[t]>>14) - (sc[0]>>14);
348 seqQ = est_seqQ(bca, types[sc[0]&0x3f], l_run);
350 tmp = sc[0]>>6 & 0xff;
351 indelQ1 = tmp > 111? 0 : (int)((1. - tmp/111.) * indelQ1 + .499); // reduce indelQ
352 sct = &score2[K*n_types];
353 for (t = 0; t < n_types; ++t) sc[t] = sct[t]<<6 | t;
354 for (t = 1; t < n_types; ++t) // insertion sort
355 for (j = t; j > 0 && sc[j] < sc[j-1]; --j)
356 tmp = sc[j], sc[j] = sc[j-1], sc[j-1] = tmp;
357 if ((sc[0]&0x3f) == ref_type) {
358 indelQ2 = (sc[1]>>14) - (sc[0]>>14);
360 for (t = 0; t < n_types; ++t) // look for the reference type
361 if ((sc[t]&0x3f) == ref_type) break;
362 indelQ2 = (sc[t]>>14) - (sc[0]>>14);
364 tmp = sc[0]>>6 & 0xff;
365 indelQ2 = tmp > 111? 0 : (int)((1. - tmp/111.) * indelQ2 + .499);
366 // pick the smaller between indelQ1 and indelQ2
367 indelQ = indelQ1 < indelQ2? indelQ1 : indelQ2;
368 p->aux = (sc[0]&0x3f)<<16 | seqQ<<8 | indelQ;
369 sumq[sc[0]&0x3f] += indelQ < seqQ? indelQ : seqQ;
370 // 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);
373 // determine bca->indel_types[] and bca->inscns
374 bca->maxins = max_ins;
375 bca->inscns = realloc(bca->inscns, bca->maxins * 4);
376 for (t = 0; t < n_types; ++t)
377 sumq[t] = sumq[t]<<6 | t;
378 for (t = 1; t < n_types; ++t) // insertion sort
379 for (j = t; j > 0 && sumq[j] > sumq[j-1]; --j)
380 tmp = sumq[j], sumq[j] = sumq[j-1], sumq[j-1] = tmp;
381 for (t = 0; t < n_types; ++t) // look for the reference type
382 if ((sumq[t]&0x3f) == ref_type) break;
383 if (t) { // then move the reference type to the first
385 for (; t > 0; --t) sumq[t] = sumq[t-1];
388 for (t = 0; t < 4; ++t) bca->indel_types[t] = B2B_INDEL_NULL;
389 for (t = 0; t < 4 && t < n_types; ++t) {
390 bca->indel_types[t] = types[sumq[t]&0x3f];
391 memcpy(&bca->inscns[t * bca->maxins], &inscns[(sumq[t]&0x3f) * max_ins], bca->maxins);
394 for (s = n_alt = 0; s < n; ++s) {
395 for (i = 0; i < n_plp[s]; ++i) {
396 bam_pileup1_t *p = plp[s] + i;
397 int x = types[p->aux>>16&0x3f];
398 for (j = 0; j < 4; ++j)
399 if (x == bca->indel_types[j]) break;
400 p->aux = j<<16 | (j == 4? 0 : (p->aux&0xffff));
401 if ((p->aux>>16&0x3f) > 0) ++n_alt;
402 // fprintf(stderr, "X pos=%d read=%d:%d name=%s call=%d type=%d q=%d seqQ=%d\n", pos, s, i, bam1_qname(p->b), p->aux>>16&63, bca->indel_types[p->aux>>16&63], p->aux&0xff, p->aux>>8&0xff);
406 free(score1); free(score2);
408 free(types); free(inscns);
409 return n_alt > 0? 0 : -1;