return cigar;
}
-/************************
- * Probabilistic glocal *
- ************************/
+/*****************************************
+ * 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 };
\ \/ \/ \/ /
\ /\ /\ /\ /
-> 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)
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] / 3.;
+ 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] = .25 * bI;
+ fi[u+0] = e * bM; fi[u+1] = EI * bI;
sum += fi[u] + fi[u+1];
}
// rescale
}
// f[2..l_query]
for (i = 2; i <= l_query; ++i) {
- double *fi = f[i], *fi1 = f[i-1], sum;
+ 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;
- // FIXME: the following line can be optimized without branching
- e = (ref[k] > 3 || query[i] > 3)? 1. : ref[k] == query[i]? 1. - qual[i] : qual[i] / 3.;
+ 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] = .25 * (m[1] * fi1[v10+0] + m[4] * fi1[v10+1]);
+ 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; k <= _end; ++k) fi[k] /= sum;
+ for (k = _beg, sum = 1./sum; k <= _end; ++k) fi[k] *= sum;
}
{ // f[l_query+1]
double sum;
// 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];
+ 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;
- // FIXME: the following can be optimized without branching
- 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];
- // FIXME: I do not know why I need this (i>1) factor, but only with it the result makes sense.
- bi[u+2] = (e * m[6] * bi1[v11+0] + m[8] * bi[v01+2]) * (i > 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
}
// rescale
set_u(_beg, bw, i, beg); set_u(_end, bw, i, end); _end += 2;
- for (k = _beg; k <= _end; ++k) bi[k] /= s[i];
+ 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] / 3.;
+ 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 + .25 * b[1][u+1] * bI;
+ sum += e * b[1][u+0] * bM + EI * b[1][u+1] * bI;
}
set_u(k, bw, 0, 0);
pb = b[0][k] = sum / s[0]; // if everything works as is expected, pb == 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;
+ 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<<2 | 0; sum += z;
- z = fi[u+1] * bi[u+1]; if (z > max) max = z, max_k = k<<2 | 1; sum += z;
+ 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;
}
max /= sum; sum *= s[i]; // 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);
+ 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, max_k>>2, max_k&3, max); // DEBUG
+ fprintf(stderr, "(%.10lg,%.10lg) (%d,%d:%d)~%lg\n", pb, sum, i-1, max_k>>2, max_k&3, max); // DEBUG
#endif
}
/*** free ***/
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!!!
+// 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);
return 0;
projects / samtools.git / blobdiff