#include <stdio.h>
#include <string.h>
#include <stdint.h>
+#include <math.h>
#include "kaln.h"
#define FROM_M 0
-2, -2, -2, -2, -2
};
-ka_param_t ka_param_blast = { 5, 2, 2, aln_sm_blast, 5, 50 };
-ka_param_t ka_param_aa2aa = { 10, 2, 2, aln_sm_blosum62, 22, 50 };
+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 };
static uint32_t *ka_path2cigar32(const path_t *path, int path_len, int *n_cigar)
{
}
#define set_end_I(II, cur, p) \
{ \
- if (gap_end >= 0) { \
- if ((p)->M - gap_open > (p)->I) { \
+ if (gap_end_ext >= 0) { \
+ if ((p)->M - gap_end_open > (p)->I) { \
(cur)->It = FROM_M; \
- (II) = (p)->M - gap_open - gap_end; \
+ (II) = (p)->M - gap_end_open - gap_end_ext; \
} else { \
(cur)->It = FROM_I; \
- (II) = (p)->I - gap_end; \
+ (II) = (p)->I - gap_end_ext; \
} \
} else set_I(II, cur, p); \
}
}
#define set_end_D(DD, cur, p) \
{ \
- if (gap_end >= 0) { \
- if ((p)->M - gap_open > (p)->D) { \
+ if (gap_end_ext >= 0) { \
+ if ((p)->M - gap_end_open > (p)->D) { \
(cur)->Dt = FROM_M; \
- (DD) = (p)->M - gap_open - gap_end; \
+ (DD) = (p)->M - gap_end_open - gap_end_ext; \
} else { \
(cur)->Dt = FROM_D; \
- (DD) = (p)->D - gap_end; \
+ (DD) = (p)->D - gap_end_ext; \
} \
} else set_D(DD, cur, p); \
}
int M, I, D;
} dpscore_t;
-/* build score profile for accelerating alignment, in theory */
-static void aln_init_score_array(uint8_t *seq, int len, int row, int *score_matrix, int **s_array)
-{
- int *tmp, *tmp2, i, k;
- for (i = 0; i != row; ++i) {
- tmp = score_matrix + i * row;
- tmp2 = s_array[i];
- for (k = 0; k != len; ++k)
- tmp2[k] = tmp[seq[k]];
- }
-}
/***************************
* banded global alignment *
***************************/
uint8_t type, ctype;
uint32_t *cigar = 0;
- int gap_open, gap_ext, gap_end, b;
+ int gap_open, gap_ext, gap_end_open, gap_end_ext, b;
int *score_matrix, N_MATRIX_ROW;
/* initialize some align-related parameters. just for compatibility */
gap_open = ap->gap_open;
gap_ext = ap->gap_ext;
- gap_end = ap->gap_end;
+ gap_end_open = ap->gap_end_open;
+ gap_end_ext = ap->gap_end_ext;
b = ap->band_width;
score_matrix = ap->matrix;
N_MATRIX_ROW = ap->row;
return cigar;
}
+
+/***************************
+ * Probabilistic extension *
+ ***************************/
+
+#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 = { 0.0001, 0.1, 10 };
+
+/*
+ The profile HMM is:
+
+ /\ /\ /\ /\ /\
+ I[0] I[1] I[k-1] I[k] I[L]
+ ^ \ ^ \ \ ^ \ ^ \ \ ^
+ | \ | \ \ | \ | \ \ |
+ M[0] -> M[1] -> ... -> M[k-1] -> M[k] -> ... -> M[L] M[L+1]
+ \ \/ \/ \/ \/ /
+ \ /\ /\ /\ /\ /
+ D[1] -> -> D[k-1] -> D[k] ->
+ \/ \/ \/
+
+ Every {M,I}[k], k=0..L connects M[L+1] at the same probability, while
+ no D[k], k=1..L-1 connects M[L+1]. This means an alignment can end up
+ with M or I but not D.
+
+ Deletions are dumb states which do not emit residues. Frankly, I am
+ not sure if they are handled properly in the following
+ implementation. This is a potential concern to be resolved in future.
+ */
+int ka_prob_extend(uint8_t *_ref, int l_ref, uint8_t *_query, int l_query, float *_qual,
+ const ka_probpar_t *c, int *state, uint8_t *q)
+{
+ double **f, **b, m[9], f_sink, sI, sM, pf, pb, pD;
+ float *qual;
+ uint8_t *ref, *query;
+ int bw, bw2, i, k, is_diff = 0;
+ /*** initialization ***/
+ ref = _ref - 1;
+ query = _query - 1;
+ qual = _qual - 1;
+ bw = c->bw;
+ bw2 = c->bw * 2 + 1;
+ 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));
+ b[i] = calloc(bw2 * 3 + 6, sizeof(double));
+ }
+ // initialize m
+ sM = sI = 1. / (2 * l_query + 2);
+ 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;
+ /*** forward ***/
+ // f[0]
+ set_u(k, bw, 0, 0);
+ f[0][k] = 1.;
+ // f[1..l_query]; core loop
+ for (i = 1; i <= l_query; ++i) {
+ double *fi = f[i], *fi1 = f[i-1];
+ int beg = 0, end = l_ref, x;
+ x = i - bw; beg = beg > x? beg : x;
+ x = i + bw; end = end < x? end : x;
+ for (k = beg; k <= end; ++k) {
+ int u, v11, v01, v10;
+ double e;
+ e = (ref[k] > 3 || query[i] > 3)? 1. : ref[k] == query[i]? 1. - qual[i] : qual[i] / 3.;
+ 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] = .25 * (m[1] * fi1[v10+0] + m[4] * fi1[v10+1]);
+ fi[u+2] = m[2] * fi[v01+0] + m[8] * fi[v01+2];
+ fprintf(stderr, "F (%d,%d;%d): %lg,%lg,%lg\n", i, k, u, fi[u], fi[u+1], fi[u+2]); // DEBUG
+ }
+ }
+ // sink
+ for (k = 0, f_sink = 0.; k <= l_ref; ++k) {
+ int u;
+ set_u(u, bw, l_query, k);
+ if (u < 3 || u >= bw2*3+3) continue;
+ f_sink += f[l_query][u+0] * sM + f[l_query][u+1] * sI;
+ }
+ pf = f_sink;
+ /*** backward ***/
+ // sink
+ for (k = 0; 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; bi[u+1] = sI;
+ }
+ // b[l_query-1..1]; core loop
+ for (i = l_query - 1; i > 0; --i) {
+ int beg = 0, end = l_ref, x;
+ double *bi = b[i], *bi1 = b[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;
+ e = k >= l_ref? 0 : (ref[k+1] > 3 || query[i+1] > 3)? 1. : ref[k+1] == query[i+1]? 1. - qual[i+1] : qual[i+1] / 3.;
+ 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);
+ bi[u+0] = e * m[0] * bi1[v11+0] + .25 * m[1] * bi1[v10+1] + m[2] * bi[v01+2];
+ bi[u+1] = e * m[3] * bi1[v11+0] + .25 * m[4] * bi1[v10+1];
+ bi[u+2] = e * m[6] * bi1[v11+0] + m[8] * bi[v01+2];
+ fprintf(stderr, "B (%d,%d;%d): %lg,%lg,%lg\n", i, k, u, bi[u], bi[u+1], bi[u+2]); // DEBUG
+ }
+ }
+ { // the "zero" coordinate (i==k==0)
+ double *bi = b[0], *bi1 = b[1];
+ int u, v11, v01, v10;
+ double e = (ref[1] > 3 || query[1] > 3)? 1. : ref[1] == query[1]? 1. - qual[1] : qual[1] / 3.;
+ set_u(u, bw, 0, 0); set_u(v11, bw, 1, 1); set_u(v10, bw, 1, 0); set_u(v01, bw, 0, 1);
+ bi[u+0] = e * m[0] * bi1[v11+0] + .25 * m[1] * bi1[v10+1] + m[2] * bi[v01+2];
+ }
+ set_u(k, bw, 0, 0);
+ pb = b[0][k];
+ pD = (pf + pb) / 2.;
+ is_diff = fabs(pb/pf - 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 = 0, 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<<2 | 0; sum += z;
+ z = fi[u+1] * bi[u+1]; if (z > max) max = z, max_k = k<<2 | 1; sum += z;
+ }
+ max /= sum; sum /= pD; // if everything works as is expected, sum == 1.0
+ if (state) state[i-1] = max_k;
+ if (q) q[i-1] = -4.343 * log(1. - max);
+#ifdef _MAIN
+ fprintf(stderr, "[%d],%d,%lg,%d:%d,%lg\n", is_diff, i, sum, max_k>>2, max_k&3, max); // DEBUG
+#endif
+ }
+ /*** free ***/
+ for (i = 0; i <= l_query; ++i) {
+ free(f[i]); free(b[i]);
+ }
+ free(f); free(b);
+ return 0;
+}
+
+#ifdef _MAIN
+int main()
+{
+ 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"; // FIXME: the output is not so right given this input!!!
+ static float qual[4] = {.01, .01, .01, .01};
+ ka_prob_extend(ref, l_ref, query, l_query, qual, &ka_probpar_def, 0, 0);
+ return 0;
+}
+#endif
projects / samtools.git / blobdiff