-2, -2, -2, -2, -2
};
+int aln_sm_qual[] = {
+ 0, -23, -23, -23, 0,
+ -23, 0, -23, -23, 0,
+ -23, -23, 0, -23, 0,
+ -23, -23, -23, 0, 0,
+ 0, 0, 0, 0, 0
+};
+
ka_param_t ka_param_blast = { 5, 2, 5, 2, aln_sm_blast, 5, 50 };
ka_param_t ka_param_aa2aa = { 10, 2, 10, 2, aln_sm_blosum62, 22, 50 };
+ka_param2_t ka_param2_qual = { 37, 11, 37, 11, 37, 11, 0, 0, aln_sm_qual, 5, 50 };
+
static uint32_t *ka_path2cigar32(const path_t *path, int path_len, int *n_cigar)
{
int i, n;
}
typedef struct {
- uint8_t Mt:3, It:2, Dt:2;
+ uint8_t Mt:3, It:2, Dt:3;
} dpcell_t;
typedef struct {
score_matrix = ap->matrix;
N_MATRIX_ROW = ap->row;
- *n_cigar = 0;
+ if (n_cigar) *n_cigar = 0;
if (len1 == 0 || len2 == 0) return 0;
/* calculate b1 and b2 */
return cigar;
}
-/*****************************************
- * Probabilistic banded glocal alignment *
- *****************************************/
-
-static float g_qual2prob[256];
-
-#define EI .25
-#define EM .3333333333333
-#define set_u(u, b, i, k) { int x=(i)-(b); x=x>0?x:0; (u)=((k)-x+1)*3; }
-
-ka_probpar_t ka_probpar_def = { 0.001, 0.1, 10 };
-
-/*
- The topology of the profile HMM:
-
- /\ /\ /\ /\
- I[1] I[k-1] I[k] I[L]
- ^ \ \ ^ \ ^ \ \ ^
- | \ \ | \ | \ \ |
- M[0] M[1] -> ... -> M[k-1] -> M[k] -> ... -> M[L] M[L+1]
- \ \/ \/ \/ /
- \ /\ /\ /\ /
- -> D[k-1] -> D[k] ->
-
- M[0] points to every {M,I}[k] and every {M,I}[k] points M[L+1].
-
- On input, _ref is the reference sequence and _query is the query
- sequence. Both are sequences of 0/1/2/3/4 where 4 stands for an
- ambiguous residue. iqual is the base quality. c sets the gap open
- probability, gap extension probability and band width.
-
- On output, state and q are arrays of length l_query. The higher 30
- bits give the reference position the query base is matched to and the
- lower two bits can be 0 (an alignment match) or 1 (an
- insertion). q[i] gives the phred scaled posterior probability of
- state[i] being wrong.
- */
-int ka_prob_glocal(const uint8_t *_ref, int l_ref, const uint8_t *_query, int l_query, const uint8_t *iqual,
- const ka_probpar_t *c, int *state, uint8_t *q)
+typedef struct {
+ int M, I, D;
+} score_aux_t;
+
+#define MINUS_INF -0x40000000
+
+// matrix: len2 rows and len1 columns
+int ka_global_score(const uint8_t *_seq1, int len1, const uint8_t *_seq2, int len2, const ka_param2_t *ap)
{
- double **f, **b, *s, m[9], sI, sM, bI, bM, pb;
- float *qual, *_qual;
- const uint8_t *ref, *query;
- int bw, bw2, i, k, is_diff = 0;
-
- /*** initialization ***/
- ref = _ref - 1; query = _query - 1; // change to 1-based coordinate
- bw = l_ref > l_query? l_ref : l_query;
- if (bw > c->bw) bw = c->bw;
- if (bw < abs(l_ref - l_query)) bw = abs(l_ref - l_query);
- bw2 = bw * 2 + 1;
- // allocate the forward and backward matrices f[][] and b[][] and the scaling array s[]
- f = calloc(l_query+1, sizeof(void*));
- b = calloc(l_query+1, sizeof(void*));
- for (i = 0; i <= l_query; ++i) {
- f[i] = calloc(bw2 * 3 + 6, sizeof(double)); // FIXME: this is over-allocated for very short seqs
- b[i] = calloc(bw2 * 3 + 6, sizeof(double));
- }
- s = calloc(l_query+2, sizeof(double)); // s[] is the scaling factor to avoid underflow
- // initialize qual
- _qual = calloc(l_query, sizeof(float));
- if (g_qual2prob[0] == 0)
- for (i = 0; i < 256; ++i)
- g_qual2prob[i] = pow(10, -i/10.);
- for (i = 0; i < l_query; ++i) _qual[i] = g_qual2prob[iqual? iqual[i] : 30];
- qual = _qual - 1;
- // initialize transition probability
- sM = sI = 1. / (2 * l_query + 2); // the value here seems not to affect results; FIXME: need proof
- m[0*3+0] = (1 - c->d - c->d) * (1 - sM); m[0*3+1] = m[0*3+2] = c->d * (1 - sM);
- m[1*3+0] = (1 - c->e) * (1 - sI); m[1*3+1] = c->e * (1 - sI); m[1*3+2] = 0.;
- m[2*3+0] = 1 - c->e; m[2*3+1] = 0.; m[2*3+2] = c->e;
- bM = (1 - c->d) / l_query; bI = c->d / l_query; // (bM+bI)*l_query==1
- /*** forward ***/
- // f[0]
- set_u(k, bw, 0, 0);
- f[0][k] = s[0] = 1.;
- { // f[1]
- double *fi = f[1], sum;
- int beg = 1, end = l_ref < bw + 1? l_ref : bw + 1, _beg, _end;
- for (k = beg, sum = 0.; k <= end; ++k) {
- int u;
- double e = (ref[k] > 3 || query[1] > 3)? 1. : ref[k] == query[1]? 1. - qual[1] : qual[1] * EM;
- set_u(u, bw, 1, k);
- fi[u+0] = e * bM; fi[u+1] = EI * bI;
- sum += fi[u] + fi[u+1];
- }
- // rescale
- s[1] = sum;
- set_u(_beg, bw, 1, beg); set_u(_end, bw, 1, end); _end += 2;
- for (k = _beg; k <= _end; ++k) fi[k] /= sum;
- }
- // f[2..l_query]
- for (i = 2; i <= l_query; ++i) {
- double *fi = f[i], *fi1 = f[i-1], sum, qli = qual[i];
- int beg = 1, end = l_ref, x, _beg, _end;
- uint8_t qyi = query[i];
- x = i - bw; beg = beg > x? beg : x; // band start
- x = i + bw; end = end < x? end : x; // band end
- for (k = beg, sum = 0.; k <= end; ++k) {
- int u, v11, v01, v10;
- double e;
- e = (ref[k] > 3 || qyi > 3)? 1. : ref[k] == qyi? 1. - qli : qli * EM;
- set_u(u, bw, i, k); set_u(v11, bw, i-1, k-1); set_u(v10, bw, i-1, k); set_u(v01, bw, i, k-1);
- fi[u+0] = e * (m[0] * fi1[v11+0] + m[3] * fi1[v11+1] + m[6] * fi1[v11+2]);
- fi[u+1] = EI * (m[1] * fi1[v10+0] + m[4] * fi1[v10+1]);
- fi[u+2] = m[2] * fi[v01+0] + m[8] * fi[v01+2];
- sum += fi[u] + fi[u+1] + fi[u+2];
-// fprintf(stderr, "F (%d,%d;%d): %lg,%lg,%lg\n", i, k, u, fi[u], fi[u+1], fi[u+2]); // DEBUG
- }
- // rescale
- s[i] = sum;
- set_u(_beg, bw, i, beg); set_u(_end, bw, i, end); _end += 2;
- for (k = _beg, sum = 1./sum; k <= _end; ++k) fi[k] *= sum;
- }
- { // f[l_query+1]
- double sum;
- for (k = 1, sum = 0.; k <= l_ref; ++k) {
- int u;
- set_u(u, bw, l_query, k);
- if (u < 3 || u >= bw2*3+3) continue;
- sum += f[l_query][u+0] * sM + f[l_query][u+1] * sI;
- }
- s[l_query+1] = sum; // the last scaling factor
+
+#define __score_aux(_p, _q0, _sc, _io, _ie, _do, _de) { \
+ int t1, t2; \
+ score_aux_t *_q; \
+ _q = _q0; \
+ _p->M = _q->M >= _q->I? _q->M : _q->I; \
+ _p->M = _p->M >= _q->D? _p->M : _q->D; \
+ _p->M += (_sc); \
+ ++_q; t1 = _q->M - _io - _ie; t2 = _q->I - _ie; _p->I = t1 >= t2? t1 : t2; \
+ _q = _p-1; t1 = _q->M - _do - _de; t2 = _q->D - _de; _p->D = t1 >= t2? t1 : t2; \
}
- /*** backward ***/
- // b[l_query] (b[l_query+1][0]=1 and thus \tilde{b}[][]=1/s[l_query+1]; this is where s[l_query+1] comes from)
- for (k = 1; k <= l_ref; ++k) {
- int u;
- double *bi = b[l_query];
- set_u(u, bw, l_query, k);
- if (u < 3 || u >= bw2*3+3) continue;
- bi[u+0] = sM / s[l_query] / s[l_query+1]; bi[u+1] = sI / s[l_query] / s[l_query+1];
+
+ int i, j, bw, scmat_size = ap->row, *scmat = ap->matrix, ret;
+ const uint8_t *seq1, *seq2;
+ score_aux_t *curr, *last, *swap;
+ bw = abs(len1 - len2) + ap->band_width;
+ i = len1 > len2? len1 : len2;
+ if (bw > i + 1) bw = i + 1;
+ seq1 = _seq1 - 1; seq2 = _seq2 - 1;
+ curr = calloc(len1 + 2, sizeof(score_aux_t));
+ last = calloc(len1 + 2, sizeof(score_aux_t));
+ { // the zero-th row
+ int x, end = len1;
+ score_aux_t *p;
+ j = 0;
+ x = j + bw; end = len1 < x? len1 : x; // band end
+ p = curr;
+ p->M = 0; p->I = p->D = MINUS_INF;
+ for (i = 1, p = &curr[1]; i <= end; ++i, ++p)
+ p->M = p->I = MINUS_INF, p->D = -(ap->edo + ap->ede * i);
+ p->M = p->I = p->D = MINUS_INF;
+ swap = curr; curr = last; last = swap;
}
- // b[l_query-1..1]
- for (i = l_query - 1; i >= 1; --i) {
- int beg = 1, end = l_ref, x, _beg, _end;
- double *bi = b[i], *bi1 = b[i+1], y = (i > 1), qli1 = qual[i+1];
- uint8_t qyi1 = query[i+1];
- x = i - bw; beg = beg > x? beg : x;
- x = i + bw; end = end < x? end : x;
- for (k = end; k >= beg; --k) {
- int u, v11, v01, v10;
- double e;
- set_u(u, bw, i, k); set_u(v11, bw, i+1, k+1); set_u(v10, bw, i+1, k); set_u(v01, bw, i, k+1);
- e = (k >= l_ref? 0 : (ref[k+1] > 3 || qyi1 > 3)? 1. : ref[k+1] == qyi1? 1. - qli1 : qli1 * EM) * bi1[v11];
- bi[u+0] = e * m[0] + EI * m[1] * bi1[v10+1] + m[2] * bi[v01+2]; // bi1[v11] has been foled into e.
- bi[u+1] = e * m[3] + EI * m[4] * bi1[v10+1];
- bi[u+2] = (e * m[6] + m[8] * bi[v01+2]) * y;
-// fprintf(stderr, "B (%d,%d;%d): %lg,%lg,%lg\n", i, k, u, bi[u], bi[u+1], bi[u+2]); // DEBUG
+ for (j = 1; j < len2; ++j) {
+ int x, beg = 0, end = len1, *scrow, col_end;
+ score_aux_t *p;
+ x = j - bw; beg = 0 > x? 0 : x; // band start
+ x = j + bw; end = len1 < x? len1 : x; // band end
+ if (beg == 0) { // from zero-th column
+ p = curr;
+ p->M = p->D = MINUS_INF; p->I = -(ap->eio + ap->eie * j);
+ ++beg; // then beg = 1
}
- // rescale
- set_u(_beg, bw, i, beg); set_u(_end, bw, i, end); _end += 2;
- for (k = _beg, y = 1./s[i]; k <= _end; ++k) bi[k] *= y;
- }
- { // b[0]
- int beg = 1, end = l_ref < bw + 1? l_ref : bw + 1;
- double sum = 0.;
- for (k = end; k >= beg; --k) {
- int u;
- double e = (ref[k] > 3 || query[1] > 3)? 1. : ref[k] == query[1]? 1. - qual[1] : qual[1] * EM;
- set_u(u, bw, 1, k);
- if (u < 3 || u >= bw2*3+3) continue;
- sum += e * b[1][u+0] * bM + EI * b[1][u+1] * bI;
+ scrow = scmat + seq2[j] * scmat_size;
+ if (end == len1) col_end = 1, --end;
+ else col_end = 0;
+ for (i = beg, p = &curr[beg]; i <= end; ++i, ++p)
+ __score_aux(p, &last[i-1], scrow[(int)seq1[i]], ap->iio, ap->iie, ap->ido, ap->ide);
+ if (col_end) {
+ __score_aux(p, &last[i-1], scrow[(int)seq1[i]], ap->eio, ap->eie, ap->ido, ap->ide);
+ ++p;
}
- set_u(k, bw, 0, 0);
- pb = b[0][k] = sum / s[0]; // if everything works as is expected, pb == 1.0
+ p->M = p->I = p->D = MINUS_INF;
+// for (i = 0; i <= len1; ++i) printf("(%d,%d,%d) ", curr[i].M, curr[i].I, curr[i].D); putchar('\n');
+ swap = curr; curr = last; last = swap;
}
- is_diff = fabs(pb - 1.) > 1e-7? 1 : 0;
- /*** MAP ***/
- for (i = 1; i <= l_query; ++i) {
- double sum = 0., *fi = f[i], *bi = b[i], max = 0.;
- int beg = 1, end = l_ref, x, max_k = -1;
- x = i - bw; beg = beg > x? beg : x;
- x = i + bw; end = end < x? end : x;
- for (k = beg; k <= end; ++k) {
- int u;
- double z;
- set_u(u, bw, i, k);
- z = fi[u+0] * bi[u+0]; if (z > max) max = z, max_k = (k-1)<<2 | 0; sum += z;
- z = fi[u+1] * bi[u+1]; if (z > max) max = z, max_k = (k-1)<<2 | 1; sum += z;
+ { // the last row
+ int x, beg = 0, *scrow;
+ score_aux_t *p;
+ j = len2;
+ x = j - bw; beg = 0 > x? 0 : x; // band start
+ if (beg == 0) { // from zero-th column
+ p = curr;
+ p->M = p->D = MINUS_INF; p->I = -(ap->eio + ap->eie * j);
+ ++beg; // then beg = 1
}
- max /= sum; sum *= s[i]; // if everything works as is expected, sum == 1.0
- if (state) state[i-1] = max_k;
- if (q) k = (int)(-4.343 * log(1. - max) + .499), q[i-1] = k > 100? 99 : k;
-#ifdef _MAIN
- fprintf(stderr, "(%.10lg,%.10lg) (%d,%d:%d)~%lg\n", pb, sum, i-1, max_k>>2, max_k&3, max); // DEBUG
-#endif
- }
- /*** free ***/
- for (i = 0; i <= l_query; ++i) {
- free(f[i]); free(b[i]);
+ scrow = scmat + seq2[j] * scmat_size;
+ for (i = beg, p = &curr[beg]; i < len1; ++i, ++p)
+ __score_aux(p, &last[i-1], scrow[(int)seq1[i]], ap->iio, ap->iie, ap->edo, ap->ede);
+ __score_aux(p, &last[i-1], scrow[(int)seq1[i]], ap->eio, ap->eie, ap->edo, ap->ede);
+// for (i = 0; i <= len1; ++i) printf("(%d,%d,%d) ", curr[i].M, curr[i].I, curr[i].D); putchar('\n');
}
- free(f); free(b); free(s); free(_qual);
- return 0;
+ ret = curr[len1].M >= curr[len1].I? curr[len1].M : curr[len1].I;
+ ret = ret >= curr[len1].D? ret : curr[len1].D;
+ free(curr); free(last);
+ return ret;
}
#ifdef _MAIN
-int main()
+int main(int argc, char *argv[])
{
- int l_ref = 5, l_query = 4;
- uint8_t *ref = (uint8_t*)"\0\1\3\3\1";
- uint8_t *query = (uint8_t*)"\0\3\3\1";
-// uint8_t *query = (uint8_t*)"\1\3\3\1";
- static uint8_t qual[4] = {20, 20, 20, 20};
- ka_prob_glocal(ref, l_ref, query, l_query, qual, &ka_probpar_def, 0, 0);
+// int len1 = 35, len2 = 35;
+// uint8_t *seq1 = (uint8_t*)"\0\0\3\3\2\0\0\0\1\0\2\1\2\1\3\2\3\3\3\0\2\3\2\1\1\3\3\3\2\3\3\1\0\0\1";
+// uint8_t *seq2 = (uint8_t*)"\0\0\3\3\2\0\0\0\1\0\2\1\2\1\3\2\3\3\3\0\2\3\2\1\1\3\3\3\2\3\3\1\0\1\0";
+ int len1 = 4, len2 = 4;
+ uint8_t *seq1 = (uint8_t*)"\1\0\0\1";
+ uint8_t *seq2 = (uint8_t*)"\1\0\1\0";
+ int sc;
+// ka_global_core(seq1, 2, seq2, 1, &ka_param_qual, &sc, 0);
+ sc = ka_global_score(seq1, len1, seq2, len2, &ka_param2_qual);
+ printf("%d\n", sc);
return 0;
}
#endif
projects / samtools.git / blobdiff