quite a lot changes to the contrast caller, but I still feel something is missing...

[samtools.git] / bcftools / prob1.c

diff --git a/bcftools/prob1.c b/bcftools/prob1.c
index afbbd419cd0242aeefd1e31776e45fecdf36a981..85c2945779f85c9cd050fd72f00cdc78f1deeacb 100644 (file)
--- a/bcftools/prob1.c
+++ b/bcftools/prob1.c
@@ -36,30 +36,42 @@ struct __bcf_p1aux_t {
        double *q2p, *pdg; // pdg -> P(D|g)
        double *phi;
        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
        const uint8_t *PL; // point to PL
        int PL_len;
 };
 
-void bcf_p1_init_prior(bcf_p1aux_t *ma, int type, double theta)
+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);
+}
+
+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;
@@ -94,10 +106,12 @@ int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn)
        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 += k * ma->phi[ma->M - k];
+       fprintf(stderr, "[heterozygosity] %lf\n", (double)sum / ma->M);
        return 0;
 }
 
-bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid
+bcf_p1aux_t *bcf_p1_init(int n)
 {
        bcf_p1aux_t *ma;
        int i;
@@ -107,6 +121,8 @@ bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid
        ma->q2p = calloc(256, sizeof(double));
        ma->pdg = calloc(3 * ma->n, sizeof(double));
        ma->phi = 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
@@ -130,7 +146,7 @@ void bcf_p1_destroy(bcf_p1aux_t *ma)
 {
        if (ma) {
                free(ma->q2p); free(ma->pdg);
-               free(ma->phi);
+               free(ma->phi); 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);
@@ -192,7 +208,7 @@ int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int 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;
 }
@@ -247,7 +263,7 @@ static void mc_cal_y(bcf_p1aux_t *ma)
 {
        if (ma->n1 > 0 && ma->n1 < ma->n) {
                int k;
-               double x;
+               long 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.;
@@ -256,26 +272,31 @@ static void mc_cal_y(bcf_p1aux_t *ma)
                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));
+               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);
-/*
-       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 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 sum1, sum2;
+       pc[0] = pc[1] = pc[2] = pc[3] = 0.;
+       if (n1 <= 0 || n2 <= 0) return;
+       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)
@@ -358,7 +379,7 @@ int bcf_p1_cal(bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
                }
        }
        rst->g[0] = rst->g[1] = rst->g[2] = -1.;
-//     bcf_p1_cal_g3(ma, rst->g);
+       contrast(ma, rst->pc);
        return 0;
 }