#include <stdlib.h>
#include <string.h>
#include <stdio.h>
+#include <errno.h>
#include "prob1.h"
+#include "kseq.h"
+KSTREAM_INIT(gzFile, gzread, 16384)
+
#define MC_AVG_ERR 0.007
#define MC_MAX_EM_ITER 16
#define MC_EM_EPS 1e-4
};
struct __bcf_p1aux_t {
- int n, M;
+ int n, M, n1;
double *q2p, *pdg; // pdg -> P(D|g)
- double *phi, *CMk; // CMk=\binom{M}{k}
+ double *phi;
double *z, *zswap; // aux for afs
+ double *z1, *z2; // only calculated when n1 is set
+ double t, t1, t2;
double *afs, *afs1; // afs: accumulative AFS; afs1: site posterior distribution
const uint8_t *PL; // point to PL
int PL_len;
}
}
+int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn)
+{
+ gzFile fp;
+ kstring_t s;
+ kstream_t *ks;
+ long double sum;
+ int dret, k;
+ memset(&s, 0, sizeof(kstring_t));
+ fp = strcmp(fn, "-")? gzopen(fn, "r") : gzdopen(fileno(stdin), "r");
+ ks = ks_init(fp);
+ memset(ma->phi, 0, sizeof(double) * (ma->M + 1));
+ while (ks_getuntil(ks, '\n', &s, &dret) >= 0) {
+ if (strstr(s.s, "[afs] ") == s.s) {
+ char *p = s.s + 6;
+ for (k = 0; k <= ma->M; ++k) {
+ int x;
+ double y;
+ x = strtol(p, &p, 10);
+ if (x != k && (errno == EINVAL || errno == ERANGE)) return -1;
+ ++p;
+ y = strtod(p, &p);
+ if (y == 0. && (errno == EINVAL || errno == ERANGE)) return -1;
+ ma->phi[ma->M - k] += y;
+ }
+ }
+ }
+ ks_destroy(ks);
+ gzclose(fp);
+ free(s.s);
+ for (sum = 0., k = 0; k <= ma->M; ++k) sum += ma->phi[k];
+ fprintf(stderr, "[prior]");
+ for (k = 0; k <= ma->M; ++k) ma->phi[k] /= sum;
+ for (k = 0; k <= ma->M; ++k) fprintf(stderr, " %d:%.3lg", k, ma->phi[ma->M - k]);
+ fputc('\n', stderr);
+ return 0;
+}
+
bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid
{
bcf_p1aux_t *ma;
int i;
ma = calloc(1, sizeof(bcf_p1aux_t));
+ ma->n1 = -1;
ma->n = n; ma->M = 2 * n;
ma->q2p = calloc(256, sizeof(double));
ma->pdg = calloc(3 * ma->n, sizeof(double));
ma->phi = calloc(ma->M + 1, sizeof(double));
- ma->CMk = calloc(ma->M + 1, sizeof(double));
ma->z = calloc(2 * ma->n + 1, sizeof(double));
ma->zswap = calloc(2 * ma->n + 1, sizeof(double));
+ ma->z1 = calloc(ma->M + 1, sizeof(double)); // actually we do not need this large
+ ma->z2 = calloc(ma->M + 1, sizeof(double));
ma->afs = calloc(2 * ma->n + 1, sizeof(double));
ma->afs1 = calloc(2 * ma->n + 1, sizeof(double));
for (i = 0; i < 256; ++i)
ma->q2p[i] = pow(10., -i / 10.);
- for (i = 0; i <= ma->M; ++i)
- ma->CMk[i] = exp(lgamma(ma->M + 1) - lgamma(i + 1) - lgamma(ma->M - i + 1));
bcf_p1_init_prior(ma, MC_PTYPE_FULL, 1e-3); // the simplest prior
return ma;
}
+int bcf_p1_set_n1(bcf_p1aux_t *b, int n1)
+{
+ if (n1 == 0 || n1 >= b->n) return -1;
+ b->n1 = n1;
+ return 0;
+}
+
void bcf_p1_destroy(bcf_p1aux_t *ma)
{
if (ma) {
free(ma->q2p); free(ma->pdg);
- free(ma->phi); free(ma->CMk);
- free(ma->z); free(ma->zswap);
+ free(ma->phi);
+ free(ma->z); free(ma->zswap); free(ma->z1); free(ma->z2);
free(ma->afs); free(ma->afs1);
free(ma);
}
return q<<2|max_i;
}
-static void mc_cal_z(bcf_p1aux_t *ma)
+#define TINY 1e-20
+
+static void mc_cal_y_core(bcf_p1aux_t *ma, int beg)
{
double *z[2], *tmp, *pdg;
- int i, j;
+ int _j, last_min, last_max;
z[0] = ma->z;
z[1] = ma->zswap;
pdg = ma->pdg;
- z[0][0] = 1.; z[0][1] = z[0][2] = 0.;
- for (j = 0; j < ma->n; ++j) {
- int max = (j + 1) * 2;
- double p[3];
- pdg = ma->pdg + j * 3;
+ memset(z[0], 0, sizeof(double) * (ma->M + 1));
+ memset(z[1], 0, sizeof(double) * (ma->M + 1));
+ z[0][0] = 1.;
+ last_min = last_max = 0;
+ ma->t = 0.;
+ for (_j = beg; _j < ma->n; ++_j) {
+ int k, j = _j - beg, _min = last_min, _max = last_max;
+ double p[3], sum;
+ pdg = ma->pdg + _j * 3;
p[0] = pdg[0]; p[1] = 2. * pdg[1]; p[2] = pdg[2];
- z[1][0] = p[0] * z[0][0];
- z[1][1] = p[0] * z[0][1] + p[1] * z[0][0];
- for (i = 2; i <= max; ++i)
- z[1][i] = p[0] * z[0][i] + p[1] * z[0][i-1] + p[2] * z[0][i-2];
- if (j < ma->n - 1) z[1][max+1] = z[1][max+2] = 0.;
+ for (; _min < _max && z[0][_min] < TINY; ++_min) z[0][_min] = z[1][_min] = 0.;
+ for (; _max > _min && z[0][_max] < TINY; --_max) z[0][_max] = z[1][_max] = 0.;
+ _max += 2;
+ if (_min == 0)
+ k = 0, z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k];
+ if (_min <= 1)
+ k = 1, z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k] + k*(2*j+2-k) * p[1] * z[0][k-1];
+ for (k = _min < 2? 2 : _min; k <= _max; ++k)
+ z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k]
+ + k*(2*j+2-k) * p[1] * z[0][k-1]
+ + k*(k-1)* p[2] * z[0][k-2];
+ for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
+ ma->t += log(sum / ((2. * j + 2) * (2. * j + 1)));
+ for (k = _min; k <= _max; ++k) z[1][k] /= sum;
+ if (_min >= 1) z[1][_min-1] = 0.;
+ if (_min >= 2) z[1][_min-2] = 0.;
+ if (j < ma->n - 1) z[1][_max+1] = z[1][_max+2] = 0.;
+ if (_j == ma->n1 - 1) { // set pop1
+ ma->t1 = ma->t;
+ memcpy(ma->z1, z[1], sizeof(double) * (ma->n1 * 2 + 1));
+ }
tmp = z[0]; z[0] = z[1]; z[1] = tmp;
+ last_min = _min; last_max = _max;
}
if (z[0] != ma->z) memcpy(ma->z, z[0], sizeof(double) * (ma->M + 1));
}
+static void mc_cal_y(bcf_p1aux_t *ma)
+{
+ if (ma->n1 > 0 && ma->n1 < ma->n) {
+ int k;
+ double x;
+ memset(ma->z1, 0, sizeof(double) * (2 * ma->n1 + 1));
+ memset(ma->z2, 0, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
+ ma->t1 = ma->t2 = 0.;
+ mc_cal_y_core(ma, ma->n1);
+ ma->t2 = ma->t;
+ memcpy(ma->z2, ma->z, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
+ mc_cal_y_core(ma, 0);
+ // rescale z
+ x = exp(ma->t - (ma->t1 + ma->t2));
+ for (k = 0; k <= ma->M; ++k) ma->z[k] *= x;
+ } else mc_cal_y_core(ma, 0);
+/*
+ if (ma->n1 > 0 && ma->n1 < ma->n) {
+ int i;
+ double y[5];
+ printf("*****\n");
+ for (i = 0; i <= 2; ++i)
+ printf("(%lf,%lf) ", ma->z1[i], ma->z2[i]);
+ printf("\n");
+ y[0] = ma->z1[0] * ma->z2[0];
+ y[1] = 1./2. * (ma->z1[0] * ma->z2[1] + ma->z1[1] * ma->z2[0]);
+ y[2] = 1./6. * (ma->z1[0] * ma->z2[2] + ma->z1[2] * ma->z2[0]) + 4./6. * ma->z1[1] * ma->z2[1];
+ y[3] = 1./2. * (ma->z1[1] * ma->z2[2] + ma->z1[2] * ma->z2[1]);
+ y[4] = ma->z1[2] * ma->z2[2];
+ for (i = 0; i <= 4; ++i) printf("(%lf,%lf) ", ma->z[i], y[i]);
+ printf("\n");
+ }
+*/
+}
+
static double mc_cal_afs(bcf_p1aux_t *ma)
{
int k;
long double sum = 0.;
memset(ma->afs1, 0, sizeof(double) * (ma->M + 1));
- mc_cal_z(ma);
+ mc_cal_y(ma);
for (k = 0, sum = 0.; k <= ma->M; ++k)
- sum += (long double)ma->phi[k] * ma->z[k] / ma->CMk[k];
+ sum += (long double)ma->phi[k] * ma->z[k];
for (k = 0; k <= ma->M; ++k) {
- ma->afs1[k] = ma->phi[k] * ma->z[k] / ma->CMk[k] / sum;
+ ma->afs1[k] = ma->phi[k] * ma->z[k] / sum;
if (isnan(ma->afs1[k]) || isinf(ma->afs1[k])) return -1.;
}
for (k = 0, sum = 0.; k <= ma->M; ++k) {
return sum / ma->M;
}
+long double bcf_p1_cal_g3(bcf_p1aux_t *p1a, double g[3])
+{
+ long double pd = 0., g2[3];
+ int i, k;
+ memset(g2, 0, sizeof(long double) * 3);
+ for (k = 0; k < p1a->M; ++k) {
+ double f = (double)k / p1a->M, f3[3], g1[3];
+ long double z = 1.;
+ g1[0] = g1[1] = g1[2] = 0.;
+ f3[0] = (1. - f) * (1. - f); f3[1] = 2. * f * (1. - f); f3[2] = f * f;
+ for (i = 0; i < p1a->n; ++i) {
+ double *pdg = p1a->pdg + i * 3;
+ double x = pdg[0] * f3[0] + pdg[1] * f3[1] + pdg[2] * f3[2];
+ z *= x;
+ g1[0] += pdg[0] * f3[0] / x;
+ g1[1] += pdg[1] * f3[1] / x;
+ g1[2] += pdg[2] * f3[2] / x;
+ }
+ pd += p1a->phi[k] * z;
+ for (i = 0; i < 3; ++i)
+ g2[i] += p1a->phi[k] * z * g1[i];
+ }
+ for (i = 0; i < 3; ++i) g[i] = g2[i] / pd;
+ return pd;
+}
+
int bcf_p1_cal(bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
{
int i, k;
rst->p_ref = ma->afs1[ma->M];
// calculate f_flat and f_em
for (k = 0, sum = 0.; k <= ma->M; ++k)
- sum += (long double)ma->z[k] / ma->CMk[k];
+ sum += (long double)ma->z[k];
rst->f_flat = 0.;
for (k = 0; k <= ma->M; ++k) {
- double p = ma->z[k] / ma->CMk[k] / sum;
+ double p = ma->z[k] / sum;
rst->f_flat += k * p;
}
rst->f_flat /= ma->M;
flast = rst->f_em;
}
}
+ rst->g[0] = rst->g[1] = rst->g[2] = -1.;
+// bcf_p1_cal_g3(ma, rst->g);
return 0;
}
projects / samtools.git / blobdiff