4 #include "bam_maqcns.h"
8 KSORT_INIT_GENERIC(uint32_t)
10 #define INDEL_WINDOW_SIZE 50
11 #define INDEL_EXT_DEP 0.9
13 typedef struct __bmc_aux_t {
21 float esum[4], fsum[4];
25 char bam_nt16_nt4_table[] = { 4, 0, 1, 4, 2, 4, 4, 4, 3, 4, 4, 4, 4, 4, 4, 4 };
28 P(<b1,b2>) = \theta \sum_{i=1}^{N-1} 1/i
29 P(D|<b1,b2>) = \sum_{k=1}^{N-1} p_k 1/2 [(k/N)^n_2(1-k/N)^n_1 + (k/N)^n1(1-k/N)^n_2]
30 p_k = 1/k / \sum_{i=1}^{N-1} 1/i
32 static void cal_het(bam_maqcns_t *aa)
35 double sum_harmo; // harmonic sum
39 aa->lhet = (double*)calloc(256 * 256, sizeof(double));
41 for (k = 1; k <= aa->n_hap - 1; ++k)
43 for (n1 = 0; n1 < 256; ++n1) {
44 for (n2 = 0; n2 < 256; ++n2) {
45 long double sum = 0.0;
46 double lC = aa->errmod == BAM_ERRMOD_SOAP? 0 : lgamma(n1+n2+1) - lgamma(n1+1) - lgamma(n2+1);
47 for (k = 1; k <= aa->n_hap - 1; ++k) {
48 double pk = 1.0 / k / sum_harmo;
49 double log1 = log((double)k/aa->n_hap);
50 double log2 = log(1.0 - (double)k/aa->n_hap);
51 sum += pk * 0.5 * (expl(log1*n2) * expl(log2*n1) + expl(log1*n1) * expl(log2*n2));
53 aa->lhet[n1<<8|n2] = lC + logl(sum);
56 poly_rate = aa->het_rate * sum_harmo;
57 aa->q_r = -4.343 * log(2.0 * poly_rate / (1.0 - poly_rate));
60 /** initialize the helper structure */
61 static void cal_coef(bam_maqcns_t *aa)
64 long double sum_a[257], b[256], q_c[256], tmp[256], fk2[256];
67 if (aa->errmod == BAM_ERRMOD_MAQ2) return; // no need to do the following
68 // aa->lhet will be allocated and initialized
69 free(aa->fk); free(aa->coef);
71 aa->fk = (double*)calloc(256, sizeof(double));
72 aa->fk[0] = fk2[0] = 1.0;
73 for (n = 1; n != 256; ++n) {
74 aa->fk[n] = pow(aa->theta, n) * (1.0 - aa->eta) + aa->eta;
75 fk2[n] = aa->fk[n>>1]; // this is an approximation, assuming reads equally likely come from both strands
77 if (aa->errmod == BAM_ERRMOD_SOAP) return;
78 aa->coef = (double*)calloc(256*256*64, sizeof(double));
79 lC = (double*)calloc(256 * 256, sizeof(double));
80 for (n = 1; n != 256; ++n)
81 for (k = 1; k <= n; ++k)
82 lC[n<<8|k] = lgamma(n+1) - lgamma(k+1) - lgamma(n-k+1);
83 for (q = 1; q != 64; ++q) {
84 double e = pow(10.0, -q/10.0);
86 double le1 = log(1.0-e);
87 for (n = 1; n != 256; ++n) {
88 double *coef = aa->coef + (q<<16|n<<8);
90 for (k = n; k >= 0; --k) { // a_k = \sum_{i=k}^n C^n_k \epsilon^k (1-\epsilon)^{n-k}
91 sum_a[k] = sum_a[k+1] + expl(lC[n<<8|k] + k*le + (n-k)*le1);
92 b[k] = sum_a[k+1] / sum_a[k];
93 if (b[k] > 0.99) b[k] = 0.99;
95 for (k = 0; k != n; ++k) // log(\bar\beta_{nk}(\bar\epsilon)^{f_k})
96 q_c[k] = -4.343 * fk2[k] * logl(b[k] / e);
97 for (k = 1; k != n; ++k) q_c[k] += q_c[k-1]; // \prod_{i=0}^k c_i
98 for (k = 0; k <= n; ++k) { // powl() in 64-bit mode seems broken on my Mac OS X 10.4.9
99 tmp[k] = -4.343 * logl(1.0 - expl(fk2[k] * logl(b[k])));
100 coef[k] = (k? q_c[k-1] : 0) + tmp[k]; // this is the final c_{nk}
107 bam_maqcns_t *bam_maqcns_init()
110 bm = (bam_maqcns_t*)calloc(1, sizeof(bam_maqcns_t));
111 bm->aux = (bmc_aux_t*)calloc(1, sizeof(bmc_aux_t));
112 bm->het_rate = 0.001;
121 void bam_maqcns_prepare(bam_maqcns_t *bm)
123 if (bm->errmod == BAM_ERRMOD_MAQ2) bm->aux->em = errmod_init(1. - bm->theta);
124 cal_coef(bm); cal_het(bm);
127 void bam_maqcns_destroy(bam_maqcns_t *bm)
130 free(bm->lhet); free(bm->fk); free(bm->coef); free(bm->aux->info); free(bm->aux->info16);
131 if (bm->aux->em) errmod_destroy(bm->aux->em);
132 free(bm->aux); free(bm);
135 glf1_t *bam_maqcns_glfgen(int _n, const bam_pileup1_t *pl, uint8_t ref_base, bam_maqcns_t *bm)
137 glf_call_aux_t *b = 0;
138 int i, j, k, w[8], c, n;
139 glf1_t *g = (glf1_t*)calloc(1, sizeof(glf1_t));
140 float p[16], min_p = 1e30;
143 g->ref_base = ref_base;
144 if (_n == 0) return g;
146 // construct aux array
147 if (bm->aux->max < _n) {
149 kroundup32(bm->aux->max);
150 bm->aux->info = (uint32_t*)realloc(bm->aux->info, 4 * bm->aux->max);
151 bm->aux->info16 = (uint16_t*)realloc(bm->aux->info16, 2 * bm->aux->max);
153 for (i = n = 0, rms = 0; i < _n; ++i) {
154 const bam_pileup1_t *p = pl + i;
155 uint32_t q, x = 0, qq;
157 if (p->is_del || p->is_refskip || (p->b->core.flag&BAM_FUNMAP)) continue;
158 q = (uint32_t)bam1_qual(p->b)[p->qpos];
159 if (q < bm->min_baseQ) continue;
160 x |= (uint32_t)bam1_strand(p->b) << 18 | q << 8 | p->b->core.qual;
161 y |= bam1_strand(p->b)<<4;
162 if (p->b->core.qual < q) q = p->b->core.qual;
163 c = p->b->core.qual < bm->cap_mapQ? p->b->core.qual : bm->cap_mapQ;
167 qq = bam1_seqi(bam1_seq(p->b), p->qpos);
168 q = bam_nt16_nt4_table[qq? qq : ref_base];
169 if (!p->is_del && !p->is_refskip && q < 4) x |= 1 << 21 | q << 16, y |= q;
170 bm->aux->info16[n] = y;
171 bm->aux->info[n++] = x;
173 rms = (uint8_t)(sqrt((double)rms / n) + .499);
174 if (bm->errmod == BAM_ERRMOD_MAQ2) {
175 errmod_cal(bm->aux->em, n, 4, bm->aux->info16, p);
178 ks_introsort(uint32_t, n, bm->aux->info);
179 // generate esum and fsum
180 b = (glf_call_aux_t*)calloc(1, sizeof(glf_call_aux_t));
181 for (k = 0; k != 8; ++k) w[k] = 0;
182 for (j = n - 1; j >= 0; --j) { // calculate esum and fsum
183 uint32_t info = bm->aux->info[j];
184 if (info>>24 < 4 && (info>>8&0x3f) != 0) info = 4<<24 | (info&0xffffff);
187 b->esum[k&3] += bm->fk[w[k]] * (info>>24);
188 b->fsum[k&3] += bm->fk[w[k]];
189 if (w[k] < 0xff) ++w[k];
194 for (j = c = 0; j != 4; ++j) c += b->c[j];
196 for (j = 0; j != 4; ++j) b->c[j] = (int)(254.0 * b->c[j] / c + 0.5);
197 for (j = c = 0; j != 4; ++j) c += b->c[j];
199 if (bm->errmod == BAM_ERRMOD_MAQ) {
200 // generate likelihood
201 for (j = 0; j != 4; ++j) {
205 for (k = 0, tmp1 = tmp3 = 0.0, tmp2 = 0; k != 4; ++k) {
206 if (j == k) continue;
207 tmp1 += b->esum[k]; tmp2 += b->c[k]; tmp3 += b->fsum[k];
210 bar_e = (int)(tmp1 / tmp3 + 0.5);
211 if (bar_e < 4) bar_e = 4; // should not happen
212 if (bar_e > 63) bar_e = 63;
213 p[j<<2|j] = tmp1 + bm->coef[bar_e<<16|c<<8|tmp2];
214 } else p[j<<2|j] = 0.0; // all the bases are j
216 for (k = j + 1; k < 4; ++k) {
217 for (i = 0, tmp2 = 0, tmp1 = tmp3 = 0.0; i != 4; ++i) {
218 if (i == j || i == k) continue;
219 tmp1 += b->esum[i]; tmp2 += b->c[i]; tmp3 += b->fsum[i];
222 bar_e = (int)(tmp1 / tmp3 + 0.5);
223 if (bar_e < 4) bar_e = 4;
224 if (bar_e > 63) bar_e = 63;
225 p[j<<2|k] = p[k<<2|j] = -4.343 * bm->lhet[b->c[j]<<8|b->c[k]] + tmp1 + bm->coef[bar_e<<16|c<<8|tmp2];
226 } else p[j<<2|k] = p[k<<2|j] = -4.343 * bm->lhet[b->c[j]<<8|b->c[k]]; // all the bases are either j or k
229 for (k = 0; k != 4; ++k)
230 if (p[j<<2|k] < 0.0) p[j<<2|k] = 0.0;
234 float max1, max2, min1, min2;
237 max1 = max2 = -1.0; min1 = min2 = 1e30;
238 for (k = 0; k < 4; ++k) {
239 if (b->esum[k] > max1) {
240 max2 = max1; max1 = b->esum[k]; max_k = k;
241 } else if (b->esum[k] > max2) max2 = b->esum[k];
243 for (k = 0; k < 4; ++k) {
244 if (p[k<<2|k] < min1) {
245 min2 = min1; min1 = p[k<<2|k]; min_k = k;
246 } else if (p[k<<2|k] < min2) min2 = p[k<<2|k];
248 if (max1 > max2 && (min_k != max_k || min1 + 1.0 > min2))
249 p[max_k<<2|max_k] = min1 > 1.0? min1 - 1.0 : 0.0;
251 } else if (bm->errmod == BAM_ERRMOD_SOAP) { // apply the SOAP model
252 // generate likelihood
253 for (j = 0; j != 4; ++j) {
256 for (k = 0, tmp = 0.0; k != 4; ++k)
257 if (j != k) tmp += b->esum[k];
260 for (k = j + 1; k < 4; ++k) {
261 for (i = 0, tmp = 0.0; i != 4; ++i)
262 if (i != j && i != k) tmp += b->esum[i];
263 p[j<<2|k] = p[k<<2|j] = -4.343 * bm->lhet[b->c[j]<<8|b->c[k]] + tmp;
269 // convert necessary information to glf1_t
270 g->ref_base = ref_base; g->max_mapQ = rms;
271 g->depth = n > 16777215? 16777215 : n;
272 for (j = 0; j != 4; ++j)
273 for (k = j; k < 4; ++k)
274 if (p[j<<2|k] < min_p) min_p = p[j<<2|k];
275 g->min_lk = min_p > 255.0? 255 : (int)(min_p + 0.5);
276 for (j = c = 0; j != 4; ++j)
277 for (k = j; k < 4; ++k)
278 g->lk[c++] = p[j<<2|k]-min_p > 255.0? 255 : (int)(p[j<<2|k]-min_p + 0.5);
284 uint32_t glf2cns(const glf1_t *g, int q_r)
286 int i, j, k, p[10], ref4;
288 ref4 = bam_nt16_nt4_table[g->ref_base];
289 for (i = k = 0; i < 4; ++i)
290 for (j = i; j < 4; ++j) {
291 int prior = (i == ref4 && j == ref4? 0 : i == ref4 || j == ref4? q_r : q_r + 3);
292 p[k] = (g->lk[k] + prior)<<4 | i<<2 | j;
295 for (i = 1; i < 10; ++i) // insertion sort
296 for (j = i; j > 0 && p[j] < p[j-1]; --j)
297 k = p[j], p[j] = p[j-1], p[j-1] = k;
298 x = (1u<<(p[0]&3) | 1u<<(p[0]>>2&3)) << 28; // the best genotype
299 x |= (uint32_t)g->max_mapQ << 16; // rms mapQ
300 x |= ((p[1]>>4) - (p[0]>>4) < 256? (p[1]>>4) - (p[0]>>4) : 255) << 8; // consensus Q
301 for (k = 0; k < 10; ++k)
302 if ((p[k]&0xf) == (ref4<<2|ref4)) break;
304 x |= (p[k]>>4) - (p[0]>>4) < 256? (p[k]>>4) - (p[0]>>4) : 255; // snp Q
308 uint32_t bam_maqcns_call(int n, const bam_pileup1_t *pl, bam_maqcns_t *bm)
313 g = bam_maqcns_glfgen(n, pl, 0xf, bm);
314 x = g->depth == 0? (0xfU<<28 | 0xfU<<24) : glf2cns(g, (int)(bm->q_r + 0.5));
316 } else x = 0xfU<<28 | 0xfU<<24;
320 /************** *****************/
322 bam_maqindel_opt_t *bam_maqindel_opt_init()
324 bam_maqindel_opt_t *mi = (bam_maqindel_opt_t*)calloc(1, sizeof(bam_maqindel_opt_t));
326 mi->r_indel = 0.00015;
335 void bam_maqindel_ret_destroy(bam_maqindel_ret_t *mir)
337 if (mir == 0) return;
338 free(mir->s[0]); free(mir->s[1]); free(mir);
341 int bam_tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, int is_left, int32_t *_tpos)
343 int k, x = c->pos, y = 0, last_y = 0;
345 for (k = 0; k < c->n_cigar; ++k) {
346 int op = cigar[k] & BAM_CIGAR_MASK;
347 int l = cigar[k] >> BAM_CIGAR_SHIFT;
348 if (op == BAM_CMATCH) {
349 if (c->pos > tpos) return y;
352 return y + (tpos - x);
356 } else if (op == BAM_CINS || op == BAM_CSOFT_CLIP) y += l;
357 else if (op == BAM_CDEL || op == BAM_CREF_SKIP) {
359 *_tpos = is_left? x : x + l;
369 #define MINUS_CONST 0x10000000
371 bam_maqindel_ret_t *bam_maqindel(int n, int pos, const bam_maqindel_opt_t *mi, const bam_pileup1_t *pl, const char *ref,
372 int _n_types, int *_types)
374 int i, j, n_types, *types, left, right, max_rd_len = 0;
375 bam_maqindel_ret_t *ret = 0;
376 // if there is no proposed indel, check if there is an indel from the alignment
378 for (i = 0; i < n; ++i) {
379 const bam_pileup1_t *p = pl + i;
380 if (!(p->b->core.flag&BAM_FUNMAP) && p->indel != 0) break;
382 if (i == n) return 0; // no indel
384 { // calculate how many types of indels are available (set n_types and types)
387 aux = (uint32_t*)calloc(n + _n_types + 1, 4);
389 aux[m++] = MINUS_CONST; // zero indel is always a type
390 for (i = 0; i < n; ++i) {
391 const bam_pileup1_t *p = pl + i;
392 if (!(p->b->core.flag&BAM_FUNMAP) && p->indel != 0)
393 aux[m++] = MINUS_CONST + p->indel;
394 j = bam_cigar2qlen(&p->b->core, bam1_cigar(p->b));
395 if (j > max_rd_len) max_rd_len = j;
397 if (_n_types) // then also add this to aux[]
398 for (i = 0; i < _n_types; ++i)
399 if (_types[i]) aux[m++] = MINUS_CONST + _types[i];
400 ks_introsort(uint32_t, m, aux);
401 // squeeze out identical types
402 for (i = 1, n_types = 1; i < m; ++i)
403 if (aux[i] != aux[i-1]) ++n_types;
404 types = (int*)calloc(n_types, sizeof(int));
406 types[j++] = aux[0] - MINUS_CONST;
407 for (i = 1; i < m; ++i) {
408 if (aux[i] != aux[i-1])
409 types[j++] = aux[i] - MINUS_CONST;
413 { // calculate left and right boundary
414 left = pos > INDEL_WINDOW_SIZE? pos - INDEL_WINDOW_SIZE : 0;
415 right = pos + INDEL_WINDOW_SIZE;
416 if (types[0] < 0) right -= types[0];
417 // in case the alignments stand out the reference
418 for (i = pos; i < right; ++i)
419 if (ref[i] == 0) break;
423 char *ref2, *rs, *inscns = 0;
424 int qr_snp, k, l, *score, *pscore, max_ins = types[n_types-1];
425 qr_snp = (int)(-4.343 * log(mi->r_snp) + .499);
426 if (max_ins > 0) { // get the consensus of inserted sequences
427 int *inscns_aux = (int*)calloc(4 * n_types * max_ins, sizeof(int));
429 for (i = 0; i < n_types; ++i) {
430 if (types[i] <= 0) continue; // not insertion
431 for (j = 0; j < n; ++j) {
432 const bam_pileup1_t *p = pl + j;
433 if (!(p->b->core.flag&BAM_FUNMAP) && p->indel == types[i]) {
434 for (k = 1; k <= p->indel; ++k) {
435 int c = bam_nt16_nt4_table[bam1_seqi(bam1_seq(p->b), p->qpos + k)];
436 if (c < 4) ++inscns_aux[i*max_ins*4 + (k-1)*4 + c];
441 // construct the consensus of inserted sequence
442 inscns = (char*)calloc(n_types * max_ins, sizeof(char));
443 for (i = 0; i < n_types; ++i) {
444 for (j = 0; j < types[i]; ++j) {
445 int max = 0, max_k = -1, *ia = inscns_aux + i*max_ins*4 + j*4;
446 for (k = 0; k < 4; ++k) {
452 inscns[i*max_ins + j] = max? 1<<max_k : 15;
458 ref2 = (char*)calloc(right - left + types[n_types-1] + 2, 1);
459 rs = (char*)calloc(right - left + max_rd_len + types[n_types-1] + 2, 1);
460 score = (int*)calloc(n_types * n, sizeof(int));
461 pscore = (int*)calloc(n_types * n, sizeof(int));
462 for (i = 0; i < n_types; ++i) {
463 ka_param_t ap = ka_param_blast;
464 ap.band_width = 2 * types[n_types - 1] + 2;
467 for (k = 0, j = left; j <= pos; ++j)
468 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
469 if (types[i] <= 0) j += -types[i];
470 else for (l = 0; l < types[i]; ++l)
471 ref2[k++] = bam_nt16_nt4_table[(int)inscns[i*max_ins + l]];
472 if (types[0] < 0) { // mask deleted sequences
473 int jj, tmp = types[i] >= 0? -types[0] : -types[0] + types[i];
474 for (jj = 0; jj < tmp && j < right && ref[j]; ++jj, ++j)
477 for (; j < right && ref[j]; ++j)
478 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
479 if (j < right) right = j;
480 // calculate score for each read
481 for (j = 0; j < n; ++j) {
482 const bam_pileup1_t *p = pl + j;
483 int qbeg, qend, tbeg, tend;
484 if (p->b->core.flag & BAM_FUNMAP) continue;
485 qbeg = bam_tpos2qpos(&p->b->core, bam1_cigar(p->b), left, 0, &tbeg);
486 qend = bam_tpos2qpos(&p->b->core, bam1_cigar(p->b), right, 1, &tend);
487 assert(tbeg >= left);
488 for (l = qbeg; l < qend; ++l)
489 rs[l - qbeg] = bam_nt16_nt4_table[bam1_seqi(bam1_seq(p->b), l)];
491 int x, y, n_acigar, ps;
494 if (tend - tbeg + types[i] <= 0) {
495 score[i*n+j] = -(1<<20);
496 pscore[i*n+j] = 1<<20;
499 acigar = ka_global_core((uint8_t*)ref2 + tbeg - left, tend - tbeg + types[i], (uint8_t*)rs, qend - qbeg, &ap, &score[i*n+j], &n_acigar);
500 x = tbeg - left; y = 0;
501 for (l = 0; l < n_acigar; ++l) {
502 int op = acigar[l]&0xf;
503 int len = acigar[l]>>4;
504 if (op == BAM_CMATCH) {
506 for (k = 0; k < len; ++k)
507 if (ref2[x+k] != rs[y+k] && ref2[x+k] < 4)
508 ps += bam1_qual(p->b)[y+k] < qr_snp? bam1_qual(p->b)[y+k] : qr_snp;
510 } else if (op == BAM_CINS || op == BAM_CSOFT_CLIP) {
511 if (op == BAM_CINS && l > 0 && l < n_acigar - 1) ps += mi->q_indel * len;
513 } else if (op == BAM_CDEL) {
514 if (l > 0 && l < n_acigar - 1) ps += mi->q_indel * len;
519 /*if (1) { // for debugging only
520 fprintf(stderr, "id=%d, pos=%d, type=%d, j=%d, score=%d, psore=%d, %d, %d, %d, %d, %d, ",
521 j, pos+1, types[i], j, score[i*n+j], pscore[i*n+j], tbeg, tend, qbeg, qend, mi->q_indel);
522 for (l = 0; l < n_acigar; ++l) fprintf(stderr, "%d%c", acigar[l]>>4, "MIDS"[acigar[l]&0xf]);
523 fprintf(stderr, "\n");
524 for (l = 0; l < tend - tbeg + types[i]; ++l) fputc("ACGTN"[ref2[l+tbeg-left]], stderr);
526 for (l = 0; l < qend - qbeg; ++l) fputc("ACGTN"[rs[l]], stderr);
533 { // get final result
534 int *sum, max1, max2, max1_i, max2_i;
535 // pick up the best two score
536 sum = (int*)calloc(n_types, sizeof(int));
537 for (i = 0; i < n_types; ++i)
538 for (j = 0; j < n; ++j)
539 sum[i] += -pscore[i*n+j];
540 max1 = max2 = -0x7fffffff; max1_i = max2_i = -1;
541 for (i = 0; i < n_types; ++i) {
543 max2 = max1; max2_i = max1_i; max1 = sum[i]; max1_i = i;
544 } else if (sum[i] > max2) {
545 max2 = sum[i]; max2_i = i;
550 ret = (bam_maqindel_ret_t*)calloc(1, sizeof(bam_maqindel_ret_t));
551 ret->indel1 = types[max1_i]; ret->indel2 = types[max2_i];
552 ret->s[0] = (char*)calloc(abs(ret->indel1) + 2, 1);
553 ret->s[1] = (char*)calloc(abs(ret->indel2) + 2, 1);
554 // write indel sequence
555 if (ret->indel1 > 0) {
557 for (k = 0; k < ret->indel1; ++k)
558 ret->s[0][k+1] = bam_nt16_rev_table[(int)inscns[max1_i*max_ins + k]];
559 } else if (ret->indel1 < 0) {
561 for (k = 0; k < -ret->indel1 && ref[pos + k + 1]; ++k)
562 ret->s[0][k+1] = ref[pos + k + 1];
563 } else ret->s[0][0] = '*';
564 if (ret->indel2 > 0) {
566 for (k = 0; k < ret->indel2; ++k)
567 ret->s[1][k+1] = bam_nt16_rev_table[(int)inscns[max2_i*max_ins + k]];
568 } else if (ret->indel2 < 0) {
570 for (k = 0; k < -ret->indel2 && ref[pos + k + 1]; ++k)
571 ret->s[1][k+1] = ref[pos + k + 1];
572 } else ret->s[1][0] = '*';
574 for (i = 0; i < n; ++i) {
575 const bam_pileup1_t *p = pl + i;
576 if (p->indel == ret->indel1) ++ret->cnt1;
577 else if (p->indel == ret->indel2) ++ret->cnt2;
578 else ++ret->cnt_anti;
581 int tmp, seq_err = 0;
583 tmp = max1_i - max2_i;
584 if (tmp < 0) tmp = -tmp;
585 for (j = 0; j < tmp + 1; ++j) x *= INDEL_EXT_DEP;
586 seq_err = mi->q_indel * (1.0 - x) / (1.0 - INDEL_EXT_DEP);
587 ret->gl[0] = ret->gl[1] = 0;
588 for (j = 0; j < n; ++j) {
589 int s1 = pscore[max1_i*n + j], s2 = pscore[max2_i*n + j];
590 //fprintf(stderr, "id=%d, %d, %d, %d, %d, %d\n", j, pl[j].b->core.pos+1, types[max1_i], types[max2_i], s1, s2);
591 if (s1 > s2) ret->gl[0] += s1 - s2 < seq_err? s1 - s2 : seq_err;
592 else ret->gl[1] += s2 - s1 < seq_err? s2 - s1 : seq_err;
595 // write cnt_ref and cnt_ambi
596 if (max1_i != 0 && max2_i != 0) {
597 for (j = 0; j < n; ++j) {
598 int diff1 = score[j] - score[max1_i * n + j];
599 int diff2 = score[j] - score[max2_i * n + j];
600 if (diff1 > 0 && diff2 > 0) ++ret->cnt_ref;
601 else if (diff1 == 0 || diff2 == 0) ++ret->cnt_ambi;
605 free(score); free(pscore); free(ref2); free(rs); free(inscns);
608 int q[3], qr_indel = (int)(-4.343 * log(mi->r_indel) + 0.5);
609 int min1, min2, min1_i;
610 q[0] = ret->gl[0] + (ret->s[0][0] != '*'? 0 : 0) * qr_indel;
611 q[1] = ret->gl[1] + (ret->s[1][0] != '*'? 0 : 0) * qr_indel;
612 q[2] = n * 3 + (ret->s[0][0] == '*' || ret->s[1][0] == '*'? 1 : 1) * qr_indel;
613 min1 = min2 = 0x7fffffff; min1_i = -1;
614 for (i = 0; i < 3; ++i) {
616 min2 = min1; min1 = q[i]; min1_i = i;
617 } else if (q[i] < min2) min2 = q[i];
620 ret->q_cns = min2 - min1;
622 if (ret->gt < 2) ret->q_ref = (ret->s[ret->gt][0] == '*')? 0 : q[1-ret->gt] - q[ret->gt] - qr_indel - 3;
623 else ret->q_ref = (ret->s[0][0] == '*')? q[0] - q[2] : q[1] - q[2];
624 if (ret->q_ref < 0) ret->q_ref = 0;