X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bcftools%2Fprob1.c;h=e3b0f7344d096f22c37048b2131219378bcb7ba5;hb=032937695157a15fb0ce042deda22afc30be1616;hp=4611af1f995fff8832300af10f733bffa41dd29d;hpb=b6f550edaa9384857879894bf01399fad76dac37;p=samtools.git diff --git a/bcftools/prob1.c b/bcftools/prob1.c index 4611af1..e3b0f73 100644 --- a/bcftools/prob1.c +++ b/bcftools/prob1.c @@ -32,32 +32,46 @@ unsigned char seq_nt4_table[256] = { }; struct __bcf_p1aux_t { - int n, M; + int n, M, n1; double *q2p, *pdg; // pdg -> P(D|g) double *phi; double *z, *zswap; // aux for afs + 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; @@ -92,20 +106,29 @@ 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 += 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); 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; 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->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 + 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) @@ -114,19 +137,24 @@ bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid 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->z); free(ma->zswap); + 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); } } -#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; @@ -180,17 +208,17 @@ 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; } #define TINY 1e-20 -static void mc_cal_y(bcf_p1aux_t *ma) +static void mc_cal_y_core(bcf_p1aux_t *ma, int beg) { double *z[2], *tmp, *pdg; - int k, j, last_min, last_max; + int _j, last_min, last_max; z[0] = ma->z; z[1] = ma->zswap; pdg = ma->pdg; @@ -198,10 +226,11 @@ static void mc_cal_y(bcf_p1aux_t *ma) memset(z[1], 0, sizeof(double) * (ma->M + 1)); z[0][0] = 1.; last_min = last_max = 0; - for (j = 0; j < ma->n; ++j) { - int _min = last_min, _max = last_max; + 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; + pdg = ma->pdg + _j * 3; p[0] = pdg[0]; p[1] = 2. * pdg[1]; p[2] = pdg[2]; 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.; @@ -215,16 +244,61 @@ static void mc_cal_y(bcf_p1aux_t *ma) + 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; + 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.; + 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 = 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); +} + +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] = -1.; + 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) { int k; @@ -244,7 +318,7 @@ static double mc_cal_afs(bcf_p1aux_t *ma) return sum / ma->M; } -static long double p1_cal_g3(bcf_p1aux_t *p1a, double g[3]) +long double bcf_p1_cal_g3(bcf_p1aux_t *p1a, double g[3]) { long double pd = 0., g2[3]; int i, k; @@ -276,7 +350,7 @@ int bcf_p1_cal(bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst) long double sum = 0.; // 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; @@ -304,7 +378,8 @@ int bcf_p1_cal(bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst) flast = rst->f_em; } } - p1_cal_g3(ma, rst->g); + rst->g[0] = rst->g[1] = rst->g[2] = -1.; + contrast(ma, rst->pc); return 0; }