#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
-
-//#define _BCF_QUAD
+#define MC_DEF_INDEL 0.15
unsigned char seq_nt4_table[256] = {
4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
};
struct __bcf_p1aux_t {
- int n, M, n1;
+ int n, M, n1, is_indel;
double *q2p, *pdg; // pdg -> P(D|g)
- double *phi;
+ double *phi, *phi_indel;
double *z, *zswap; // aux for afs
- double *z1, *z2; // only calculated when n1 is set
+ double *z1, *z2, *phi1, *phi2; // only calculated when n1 is set
double t, t1, t2;
double *afs, *afs1; // afs: accumulative AFS; afs1: site posterior distribution
- double *k1k2;
const uint8_t *PL; // point to PL
int PL_len;
};
-void bcf_p1_init_prior(bcf_p1aux_t *ma, int type, double theta)
+void bcf_p1_indel_prior(bcf_p1aux_t *ma, double x)
+{
+ int i;
+ for (i = 0; i < ma->M; ++i)
+ ma->phi_indel[i] = ma->phi[i] * x;
+ ma->phi_indel[ma->M] = 1. - ma->phi[ma->M] * x;
+}
+
+static void init_prior(int type, double theta, int M, double *phi)
{
int i;
if (type == MC_PTYPE_COND2) {
- for (i = 0; i <= ma->M; ++i)
- ma->phi[i] = 2. * (i + 1) / (ma->M + 1) / (ma->M + 2);
+ for (i = 0; i <= M; ++i)
+ phi[i] = 2. * (i + 1) / (M + 1) / (M + 2);
} else if (type == MC_PTYPE_FLAT) {
- for (i = 0; i <= ma->M; ++i)
- ma->phi[i] = 1. / (ma->M + 1);
+ for (i = 0; i <= M; ++i)
+ phi[i] = 1. / (M + 1);
} else {
double sum;
- for (i = 0, sum = 0.; i < ma->M; ++i)
- sum += (ma->phi[i] = theta / (ma->M - i));
- ma->phi[ma->M] = 1. - sum;
+ for (i = 0, sum = 0.; i < M; ++i)
+ sum += (phi[i] = theta / (M - i));
+ phi[M] = 1. - sum;
}
}
+void bcf_p1_init_prior(bcf_p1aux_t *ma, int type, double theta)
+{
+ init_prior(type, theta, ma->M, ma->phi);
+ bcf_p1_indel_prior(ma, MC_DEF_INDEL);
+}
+
+void bcf_p1_init_subprior(bcf_p1aux_t *ma, int type, double theta)
+{
+ if (ma->n1 <= 0 || ma->n1 >= ma->M) return;
+ init_prior(type, theta, 2*ma->n1, ma->phi1);
+ init_prior(type, theta, 2*(ma->n - ma->n1), ma->phi2);
+}
+
int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn)
{
gzFile fp;
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);
+ for (sum = 0., k = 1; k < ma->M; ++k) sum += ma->phi[ma->M - k] * (2.* k * (ma->M - k) / ma->M / (ma->M - 1));
+ fprintf(stderr, "[%s] heterozygosity=%lf, ", __func__, (double)sum);
+ for (sum = 0., k = 1; k <= ma->M; ++k) sum += k * ma->phi[ma->M - k] / ma->M;
+ fprintf(stderr, "theta=%lf\n", (double)sum);
+ bcf_p1_indel_prior(ma, MC_DEF_INDEL);
return 0;
}
ma->q2p = calloc(256, sizeof(double));
ma->pdg = calloc(3 * ma->n, sizeof(double));
ma->phi = calloc(ma->M + 1, sizeof(double));
+ ma->phi_indel = calloc(ma->M + 1, sizeof(double));
+ ma->phi1 = calloc(ma->M + 1, sizeof(double));
+ ma->phi2 = 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
return ma;
}
-#ifdef _BCF_QUAD
-static double lbinom(int n, int k)
-{
- return lgamma(n+1) - lgamma(k+1) - lgamma(n-k+1);
-}
-#endif
-
int bcf_p1_set_n1(bcf_p1aux_t *b, int n1)
{
if (n1 == 0 || n1 >= b->n) return -1;
b->n1 = n1;
-#ifdef _BCF_QUAD
- {
- int k1, k2, n2 = b->n - b->n1;
- b->k1k2 = calloc((2*n1+1) * (2*n2+1), sizeof(double));
- for (k1 = 0; k1 <= 2*n1; ++k1)
- for (k2 = 0; k2 <= 2*n2; ++k2)
- b->k1k2[k1*(2*n2+1)+k2] = exp(lbinom(2*n1,k1) + lbinom(2*n2,k2) - lbinom(b->M,k1+k2));
- }
-#endif
return 0;
}
{
if (ma) {
free(ma->q2p); free(ma->pdg);
- free(ma->phi);
+ free(ma->phi); free(ma->phi_indel); free(ma->phi1); free(ma->phi2);
free(ma->z); free(ma->zswap); free(ma->z1); free(ma->z2);
free(ma->afs); free(ma->afs1);
- free(ma->k1k2);
free(ma);
}
}
-#define char2int(s) (((int)s[0])<<8|s[1])
-
static int cal_pdg(const bcf1_t *b, bcf_p1aux_t *ma)
{
int i, j, k;
}
max = 1. - max;
if (max < 1e-308) max = 1e-308;
- q = (int)(-3.434 * log(max) + .499);
+ q = (int)(-4.343 * log(max) + .499);
if (q > 99) q = 99;
return q<<2|max_i;
}
x = expl(ma->t - (ma->t1 + ma->t2));
for (k = 0; k <= ma->M; ++k) ma->z[k] *= x;
} else mc_cal_y_core(ma, 0);
-#ifdef _BCF_QUAD
-/*
- if (ma->n1 > 0 && ma->n1 < ma->n) { // DEBUG: consistency check; z[i] should equal y[i]
- int i, k1, k2, n1 = ma->n1, n2 = ma->n - n1;
- double *y;
- printf("*** ");
- y = calloc(ma->M + 1, sizeof(double));
- for (k1 = 0; k1 <= 2*n1; ++k1)
- for (k2 = 0; k2 <= 2*n2; ++k2)
- y[k1+k2] += ma->k1k2[k1*(2*n2+1)+k2] * ma->z1[k1] * ma->z2[k2];
- for (i = 0; i <= ma->M; ++i) printf("(%lf,%lf) ", ma->z[i], y[i]);
- printf("\n");
- free(y);
- }
-*/
-#endif
}
static void contrast(bcf_p1aux_t *ma, double pc[4]) // mc_cal_y() must be called before hand
{
int k, n1 = ma->n1, n2 = ma->n - ma->n1;
- long double sum = -1., x, sum_alt;
- double y;
+ long double sum1, sum2;
pc[0] = pc[1] = pc[2] = pc[3] = -1.;
if (n1 <= 0 || n2 <= 0) return;
-#ifdef _BCF_QUAD
- { // FIXME: can be improved by skipping zero cells
- int k1, k2;
- long double z[3];
- z[0] = z[1] = z[2] = 0.;
- for (k1 = 0; k1 <= 2*n1; ++k1)
- for (k2 = 0; k2 <= 2*n2; ++k2) {
- double zz = ma->phi[k1+k2] * ma->z1[k1] * ma->z2[k2] * ma->k1k2[k1*(2*n2+1)+k2];
- if ((double)k1/n1 < (double)k2/n2) z[0] += zz;
- else if ((double)k1/n1 > (double)k2/n2) z[1] += zz;
- else z[2] += zz;
- }
- sum = z[0] + z[1] + z[2];
- pc[2] = z[0] / sum; pc[3] = z[1] / sum;
- }
-#endif
- for (k = 0, sum_alt = 0.; k <= ma->M; ++k)
- sum_alt += (long double)ma->phi[k] * ma->z[k];
-// printf("* %lg, %lg *\n", (double)sum, (double)sum_alt); // DEBUG: sum should equal sum_alt
- sum = sum_alt;
- y = lgamma(2*n1 + 1) - lgamma(ma->M + 1);
- for (k = 1, x = 0.; k <= 2 * n2; ++k)
- x += ma->phi[k] * ma->z2[k] * exp(lgamma(ma->M - k + 1) - lgamma(2*n2 - k + 1) + y);
- pc[0] = ma->z1[0] * x / sum;
- y = lgamma(2*n2 + 1) - lgamma(ma->M + 1);
- for (k = 1, x = 0.; k <= 2 * n1; ++k)
- x += ma->phi[k] * ma->z1[k] * exp(lgamma(ma->M - k + 1) - lgamma(2*n1 - k + 1) + y);
- pc[1] = ma->z2[0] * x / sum;
- for (k = 0; k < 4; ++k) {
- y = 1. - pc[k];
- if (y <= 0.) y = 1e-100;
- pc[k] = (int)(-3.434 * log(y) + .499);
- if (pc[k] > 99.) pc[k] = 99.;
+ for (k = 0, sum1 = 0.; k <= 2*n1; ++k) sum1 += ma->phi1[k] * ma->z1[k];
+ for (k = 0, sum2 = 0.; k <= 2*n2; ++k) sum2 += ma->phi2[k] * ma->z2[k];
+ pc[2] = ma->phi1[2*n1] * ma->z1[2*n1] / sum1;
+ pc[3] = ma->phi2[2*n2] * ma->z2[2*n2] / sum2;
+ for (k = 2; k < 4; ++k) {
+ pc[k] = pc[k] > .5? -(-4.343 * log(1. - pc[k] + TINY) + .499) : -4.343 * log(pc[k] + TINY) + .499;
+ pc[k] = (int)pc[k];
+ if (pc[k] > 99) pc[k] = 99;
+ if (pc[k] < -99) pc[k] = -99;
}
+ pc[0] = ma->phi2[2*n2] * ma->z2[2*n2] / sum2 * (1. - ma->phi1[2*n1] * ma->z1[2*n1] / sum1);
+ pc[1] = ma->phi1[2*n1] * ma->z1[2*n1] / sum1 * (1. - ma->phi2[2*n2] * ma->z2[2*n2] / sum2);
+ pc[0] = pc[0] == 1.? 99 : (int)(-4.343 * log(1. - pc[0]) + .499);
+ pc[1] = pc[1] == 1.? 99 : (int)(-4.343 * log(1. - pc[1]) + .499);
}
-static double mc_cal_afs(bcf_p1aux_t *ma)
+static double mc_cal_afs(bcf_p1aux_t *ma, double *p_ref_folded, double *p_var_folded)
{
int k;
- long double sum = 0.;
+ long double sum = 0., sum2;
+ double *phi = ma->is_indel? ma->phi_indel : ma->phi;
memset(ma->afs1, 0, sizeof(double) * (ma->M + 1));
mc_cal_y(ma);
+ // compute AFS
for (k = 0, sum = 0.; k <= ma->M; ++k)
- sum += (long double)ma->phi[k] * ma->z[k];
+ sum += (long double)phi[k] * ma->z[k];
for (k = 0; k <= ma->M; ++k) {
- ma->afs1[k] = ma->phi[k] * ma->z[k] / sum;
+ ma->afs1[k] = phi[k] * ma->z[k] / sum;
if (isnan(ma->afs1[k]) || isinf(ma->afs1[k])) return -1.;
}
+ // compute folded variant probability
+ for (k = 0, sum = 0.; k <= ma->M; ++k)
+ sum += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
+ for (k = 1, sum2 = 0.; k < ma->M; ++k)
+ sum2 += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
+ *p_var_folded = sum2 / sum;
+ *p_ref_folded = (phi[k] + phi[ma->M - k]) / 2. * (ma->z[ma->M] + ma->z[0]) / sum;
+ // the expected frequency
for (k = 0, sum = 0.; k <= ma->M; ++k) {
ma->afs[k] += ma->afs1[k];
sum += k * ma->afs1[k];
return pd;
}
-int bcf_p1_cal(bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
+int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
{
int i, k;
long double sum = 0.;
+ ma->is_indel = bcf_is_indel(b);
// set PL and PL_len
for (i = 0; i < b->n_gi; ++i) {
- if (b->gi[i].fmt == char2int("PL")) {
+ if (b->gi[i].fmt == bcf_str2int("PL", 2)) {
ma->PL = (uint8_t*)b->gi[i].data;
ma->PL_len = b->gi[i].len;
break;
if (b->n_alleles < 2) return -1; // FIXME: find a better solution
//
rst->rank0 = cal_pdg(b, ma);
- rst->f_exp = mc_cal_afs(ma);
+ rst->f_exp = mc_cal_afs(ma, &rst->p_ref_folded, &rst->p_var_folded);
rst->p_ref = ma->afs1[ma->M];
+ for (k = 0, sum = 0.; k < ma->M; ++k)
+ sum += ma->afs1[k];
+ rst->p_var = (double)sum;
// calculate f_flat and f_em
for (k = 0, sum = 0.; k <= ma->M; ++k)
sum += (long double)ma->z[k];
flast = rst->f_em;
}
}
+ { // estimate equal-tail credible interval (95% level)
+ int l, h;
+ double p;
+ for (i = 0, p = 0.; i < ma->M; ++i)
+ if (p + ma->afs1[i] > 0.025) break;
+ else p += ma->afs1[i];
+ l = i;
+ for (i = ma->M-1, p = 0.; i >= 0; --i)
+ if (p + ma->afs1[i] > 0.025) break;
+ else p += ma->afs1[i];
+ h = i;
+ rst->cil = (double)(ma->M - h) / ma->M; rst->cih = (double)(ma->M - l) / ma->M;
+ }
rst->g[0] = rst->g[1] = rst->g[2] = -1.;
contrast(ma, rst->pc);
-// bcf_p1_cal_g3(ma, rst->g);
return 0;
}
projects / samtools.git / blobdiff