7 #define MINUS_CONST 0x10000000
8 #define INDEL_WINDOW_SIZE 50
10 static int tpos2qpos(const bam1_core_t *c, const uint32_t *cigar, int32_t tpos, int is_left, int32_t *_tpos)
12 int k, x = c->pos, y = 0, last_y = 0;
14 for (k = 0; k < c->n_cigar; ++k) {
15 int op = cigar[k] & BAM_CIGAR_MASK;
16 int l = cigar[k] >> BAM_CIGAR_SHIFT;
17 if (op == BAM_CMATCH) {
18 if (c->pos > tpos) return y;
21 return y + (tpos - x);
25 } else if (op == BAM_CINS || op == BAM_CSOFT_CLIP) y += l;
26 else if (op == BAM_CDEL || op == BAM_CREF_SKIP) {
28 *_tpos = is_left? x : x + l;
37 // FIXME: check if the inserted sequence is consistent with the homopolymer run
38 // l is the relative gap length and l_run is the length of the homopolymer on the reference
39 static inline int est_seqQ(const bcf_callaux_t *bca, int l, int l_run)
42 q = bca->openQ + bca->extQ * (abs(l) - 1);
43 qh = l_run >= 3? (int)(bca->tandemQ * (double)abs(l) / l_run + .499) : 1000;
44 return q < qh? q : qh;
47 int bcf_call_gap_prep(int n, int *n_plp, bam_pileup1_t **plp, int pos, bcf_callaux_t *bca, const char *ref)
49 extern void ks_introsort_uint32_t(int, uint32_t*);
50 int i, s, j, k, t, n_types, *types, max_rd_len, left, right, max_ins, *score, N, K, l_run, ref_type;
51 char *inscns = 0, *ref2, *query;
52 if (ref == 0 || bca == 0) return -1;
53 // determine if there is a gap
54 for (s = N = 0; s < n; ++s) {
55 N += n_plp[s]; // N is the total number of reads
56 for (i = 0; i < n_plp[s]; ++i)
57 if (plp[s][i].indel != 0) break;
58 if (i < n_plp[s]) break;
60 if (s == n) return -1; // there is no indel at this position.
61 { // find out how many types of indels are present
64 aux = calloc(N + 1, 4);
66 aux[m++] = MINUS_CONST; // zero indel is always a type
67 for (s = 0; s < n; ++s) {
68 for (i = 0; i < n_plp[s]; ++i) {
69 const bam_pileup1_t *p = plp[s] + i;
71 aux[m++] = MINUS_CONST + p->indel;
72 j = bam_cigar2qlen(&p->b->core, bam1_cigar(p->b));
73 if (j > max_rd_len) max_rd_len = j;
76 ks_introsort(uint32_t, m, aux);
77 // squeeze out identical types
78 for (i = 1, n_types = 1; i < m; ++i)
79 if (aux[i] != aux[i-1]) ++n_types;
80 assert(n_types > 1); // there must at least one type of non-reference indel
81 types = (int*)calloc(n_types, sizeof(int));
83 types[t++] = aux[0] - MINUS_CONST;
84 for (i = 1; i < m; ++i)
85 if (aux[i] != aux[i-1])
86 types[t++] = aux[i] - MINUS_CONST;
88 for (t = 0; t < n_types; ++t)
89 if (types[t] == 0) break;
90 ref_type = t; // the index of the reference type (0)
93 { // calculate left and right boundary
94 left = pos > INDEL_WINDOW_SIZE? pos - INDEL_WINDOW_SIZE : 0;
95 right = pos + INDEL_WINDOW_SIZE;
96 if (types[0] < 0) right -= types[0];
97 // in case the alignments stand out the reference
98 for (i = pos; i < right; ++i)
99 if (ref[i] == 0) break;
102 { // the length of the homopolymer run around the current position
103 int c = bam_nt16_table[(int)ref[pos + 1]];
104 if (c == 15) l_run = 1;
106 for (i = pos + 2; ref[i]; ++i)
107 if (bam_nt16_table[(int)ref[i]] != c) break;
109 for (i = pos; i >= 0; --i)
110 if (bam_nt16_table[(int)ref[i]] != c) break;
114 // construct the consensus sequence
115 max_ins = types[n_types - 1]; // max_ins is at least 0
117 int *inscns_aux = calloc(4 * n_types * max_ins, sizeof(int));
118 // count the number of occurrences of each base at each position for each type of insertion
119 for (t = 0; t < n_types; ++t) {
121 for (s = 0; s < n; ++s) {
122 for (i = 0; i < n_plp[s]; ++i) {
123 bam_pileup1_t *p = plp[s] + i;
124 if (p->indel == types[t]) {
125 uint8_t *seq = bam1_seq(p->b);
126 for (k = 1; k <= p->indel; ++k) {
127 int c = bam_nt16_nt4_table[bam1_seqi(seq, p->qpos + k)];
128 if (c < 4) ++inscns_aux[(t*max_ins+(k-1))*4 + c];
135 // use the majority rule to construct the consensus
136 inscns = calloc(n_types * max_ins, 1);
137 for (t = 0; t < n_types; ++t) {
138 for (j = 0; j < types[t]; ++j) {
139 int max = 0, max_k = -1, *ia = &inscns_aux[(t*max_ins+j)*4];
140 for (k = 0; k < 4; ++k)
142 max = ia[k], max_k = k;
143 inscns[t*max_ins + j] = max? max_k : 4;
148 // compute the likelihood given each type of indel for each read
149 ref2 = calloc(right - left + max_ins + 2, 1);
150 query = calloc(right - left + max_rd_len + max_ins + 2, 1);
151 score = calloc(N * n_types, sizeof(int));
152 for (t = 0; t < n_types; ++t) {
154 ka_param2_t ap = ka_param2_qual;
155 ap.band_width = abs(types[t]) + 3;
157 for (k = 0, j = left; j <= pos; ++j)
158 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
159 if (types[t] <= 0) j += -types[t];
160 else for (l = 0; l < types[t]; ++l)
161 ref2[k++] = inscns[t*max_ins + l];
162 if (types[0] < 0) { // mask deleted sequences to avoid a particular error in the model.
163 int jj, tmp = types[t] >= 0? -types[0] : -types[0] + types[t];
164 for (jj = 0; jj < tmp && j < right && ref[j]; ++jj, ++j)
167 for (; j < right && ref[j]; ++j)
168 ref2[k++] = bam_nt16_nt4_table[bam_nt16_table[(int)ref[j]]];
169 if (j < right) right = j;
170 // align each read to ref2
171 for (s = K = 0; s < n; ++s) {
172 for (i = 0; i < n_plp[s]; ++i, ++K) {
173 bam_pileup1_t *p = plp[s] + i;
174 int qbeg, qend, tbeg, tend, sc;
175 uint8_t *seq = bam1_seq(p->b);
176 // determine the start and end of sequences for alignment
177 qbeg = tpos2qpos(&p->b->core, bam1_cigar(p->b), left, 0, &tbeg);
178 qend = tpos2qpos(&p->b->core, bam1_cigar(p->b), right, 1, &tend);
181 tbeg = tbeg - l > left? tbeg - l : left;
183 // write the query sequence
184 for (l = qbeg; l < qend; ++l)
185 query[l - qbeg] = bam_nt16_nt4_table[bam1_seqi(seq, l)];
186 // do alignment; this takes most of computing time for indel calling
187 sc = ka_global_score((uint8_t*)ref2 + tbeg - left, tend - tbeg + abs(types[t]),
188 (uint8_t*)query + qbeg, qend - qbeg, &ap);
189 score[K*n_types + t] = -sc;
191 for (l = 0; l < tend - tbeg + abs(types[t]); ++l)
192 fputc("ACGTN"[(int)ref2[tbeg-left+l]], stderr);
194 for (l = 0; l < qend - qbeg; ++l) fputc("ACGTN"[(int)query[qbeg + l]], stderr);
196 fprintf(stderr, "pos=%d type=%d read=%d:%d name=%s score=%d\n", pos, types[t], s, i, bam1_qname(p->b), sc);
201 free(ref2); free(query);
204 sc = alloca(n_types * sizeof(int));
205 sumq = alloca(n_types * sizeof(int));
206 memset(sumq, 0, sizeof(int) * n_types);
207 for (s = K = 0; s < n; ++s) {
208 for (i = 0; i < n_plp[s]; ++i, ++K) {
209 bam_pileup1_t *p = plp[s] + i;
210 int *sct = &score[K*n_types], indelQ;
211 for (t = 0; t < n_types; ++t) sc[t] = sct[t]<<6 | t;
212 for (t = 1; t < n_types; ++t) // insertion sort
213 for (j = t; j > 0 && sc[j] < sc[j-1]; --j)
214 tmp = sc[j], sc[j] = sc[j-1], sc[j-1] = tmp;
215 /* errmod_cal() assumes that if the call is wrong, the
216 * likelihoods of other events are equal. This is about
217 * right for substitutions, but is not desired for
218 * indels. To reuse errmod_cal(), I have to make
219 * compromise for multi-allelic indels.
221 if ((sc[0]&0x3f) == ref_type) {
222 indelQ = (sc[1]>>6) - (sc[0]>>6);
223 tmp = est_seqQ(bca, types[sc[1]&0x3f], l_run);
225 for (t = 0; t < n_types; ++t) // look for the reference type
226 if ((sc[t]&0x3f) == ref_type) break;
227 indelQ = (sc[t]>>6) - (sc[0]>>6);
228 tmp = est_seqQ(bca, types[sc[0]&0x3f], l_run);
230 if (indelQ > tmp) indelQ = tmp;
231 p->aux = (sc[0]&0x3f)<<8 | indelQ;
232 sumq[sc[0]&0x3f] += indelQ;
233 // 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);
236 // determine bca->indel_types[]
237 for (t = 0; t < n_types; ++t)
238 sumq[t] = sumq[t]<<6 | t;
239 for (t = 1; t < n_types; ++t) // insertion sort
240 for (j = t; j > 0 && sumq[j] < sumq[j-1]; --j)
241 tmp = sumq[j], sumq[j] = sumq[j-1], sumq[j-1] = tmp;
242 for (t = 0; t < n_types; ++t) // look for the reference type
243 if ((sumq[t]&0x3f) == ref_type) break;
244 if (t) { // then move the reference type to the first
246 for (; t > 0; --t) sumq[t] = sumq[t-1];
249 for (t = 0; t < 4; ++t) bca->indel_types[t] = B2B_INDEL_NULL;
250 for (t = 0; t < 4 && t < n_types; ++t)
251 bca->indel_types[t] = types[sumq[t]&0x3f];
253 for (s = K = 0; s < n; ++s) {
254 for (i = 0; i < n_plp[s]; ++i, ++K) {
255 bam_pileup1_t *p = plp[s] + i;
256 int x = types[p->aux>>8&0x3f];
257 for (j = 0; j < 4; ++j)
258 if (x == bca->indel_types[j]) break;
259 p->aux = j<<8 | (j == 4? 0 : (p->aux&0xff));
260 // fprintf(stderr, "pos=%d read=%d:%d name=%s call=%d q=%d\n", pos, s, i, bam1_qname(p->b), p->aux>>8&63, p->aux&0xff);
264 // FIXME: to set the inserted sequence
267 free(types); free(inscns);