X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bcftools%2Fprob1.c;h=1d0328f936c180ebb447e4006631fa2c018363b6;hb=b62e611309f1aab8113ae4f0f74b0a61113f3cda;hp=e124b0f8f7810114f47c5d7eed3178681849877c;hpb=082c52f3cb5517db50987bf1dc43aef845c45fd8;p=samtools.git diff --git a/bcftools/prob1.c b/bcftools/prob1.c index e124b0f..1d0328f 100644 --- a/bcftools/prob1.c +++ b/bcftools/prob1.c @@ -12,6 +12,8 @@ KSTREAM_INIT(gzFile, gzread, 16384) #define MC_MAX_EM_ITER 16 #define MC_EM_EPS 1e-4 +//#define _BCF_QUAD + unsigned char seq_nt4_table[256] = { 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, @@ -36,7 +38,10 @@ 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 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; }; @@ -95,7 +100,7 @@ int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn) 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 +112,8 @@ bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid ma->phi = 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) @@ -115,10 +122,26 @@ bcf_p1aux_t *bcf_p1_init(int n) // FIXME: assuming diploid 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; } @@ -127,8 +150,9 @@ 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->z); free(ma->zswap); free(ma->z1); free(ma->z2); free(ma->afs); free(ma->afs1); + free(ma->k1k2); free(ma); } } @@ -195,10 +219,10 @@ int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int k) #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; @@ -206,10 +230,11 @@ 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.; @@ -223,16 +248,100 @@ 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); +#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; + 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; + } +#else + pc[2] = pc[3] = 0.; +#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.; + } +} + static double mc_cal_afs(bcf_p1aux_t *ma) { int k; @@ -252,7 +361,7 @@ 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; @@ -312,7 +421,9 @@ 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); +// bcf_p1_cal_g3(ma, rst->g); return 0; }