10 KSTREAM_INIT(gzFile, gzread, 16384)
12 #define MC_MAX_EM_ITER 16
13 #define MC_EM_EPS 1e-5
14 #define MC_DEF_INDEL 0.15
16 unsigned char seq_nt4_table[256] = {
17 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
18 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
19 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4 /*'-'*/, 4, 4,
20 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
21 4, 0, 4, 1, 4, 4, 4, 2, 4, 4, 4, 4, 4, 4, 4, 4,
22 4, 4, 4, 4, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
23 4, 0, 4, 1, 4, 4, 4, 2, 4, 4, 4, 4, 4, 4, 4, 4,
24 4, 4, 4, 4, 3, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
25 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
26 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
27 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
28 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
29 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
30 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
31 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
32 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4
35 struct __bcf_p1aux_t {
36 int n, M, n1, is_indel;
37 uint8_t *ploidy; // haploid or diploid ONLY
38 double *q2p, *pdg; // pdg -> P(D|g)
39 double *phi, *phi_indel;
40 double *z, *zswap; // aux for afs
41 double *z1, *z2, *phi1, *phi2; // only calculated when n1 is set
42 double **hg; // hypergeometric distribution
44 double *afs, *afs1; // afs: accumulative AFS; afs1: site posterior distribution
45 const uint8_t *PL; // point to PL
49 void bcf_p1_indel_prior(bcf_p1aux_t *ma, double x)
52 for (i = 0; i < ma->M; ++i)
53 ma->phi_indel[i] = ma->phi[i] * x;
54 ma->phi_indel[ma->M] = 1. - ma->phi[ma->M] * x;
57 static void init_prior(int type, double theta, int M, double *phi)
60 if (type == MC_PTYPE_COND2) {
61 for (i = 0; i <= M; ++i)
62 phi[i] = 2. * (i + 1) / (M + 1) / (M + 2);
63 } else if (type == MC_PTYPE_FLAT) {
64 for (i = 0; i <= M; ++i)
65 phi[i] = 1. / (M + 1);
68 for (i = 0, sum = 0.; i < M; ++i)
69 sum += (phi[i] = theta / (M - i));
74 void bcf_p1_init_prior(bcf_p1aux_t *ma, int type, double theta)
76 init_prior(type, theta, ma->M, ma->phi);
77 bcf_p1_indel_prior(ma, MC_DEF_INDEL);
80 void bcf_p1_init_subprior(bcf_p1aux_t *ma, int type, double theta)
82 if (ma->n1 <= 0 || ma->n1 >= ma->M) return;
83 init_prior(type, theta, 2*ma->n1, ma->phi1);
84 init_prior(type, theta, 2*(ma->n - ma->n1), ma->phi2);
87 int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn)
94 memset(&s, 0, sizeof(kstring_t));
95 fp = strcmp(fn, "-")? gzopen(fn, "r") : gzdopen(fileno(stdin), "r");
97 memset(ma->phi, 0, sizeof(double) * (ma->M + 1));
98 while (ks_getuntil(ks, '\n', &s, &dret) >= 0) {
99 if (strstr(s.s, "[afs] ") == s.s) {
101 for (k = 0; k <= ma->M; ++k) {
104 x = strtol(p, &p, 10);
105 if (x != k && (errno == EINVAL || errno == ERANGE)) return -1;
108 if (y == 0. && (errno == EINVAL || errno == ERANGE)) return -1;
109 ma->phi[ma->M - k] += y;
116 for (sum = 0., k = 0; k <= ma->M; ++k) sum += ma->phi[k];
117 fprintf(stderr, "[prior]");
118 for (k = 0; k <= ma->M; ++k) ma->phi[k] /= sum;
119 for (k = 0; k <= ma->M; ++k) fprintf(stderr, " %d:%.3lg", k, ma->phi[ma->M - k]);
121 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));
122 fprintf(stderr, "[%s] heterozygosity=%lf, ", __func__, (double)sum);
123 for (sum = 0., k = 1; k <= ma->M; ++k) sum += k * ma->phi[ma->M - k] / ma->M;
124 fprintf(stderr, "theta=%lf\n", (double)sum);
125 bcf_p1_indel_prior(ma, MC_DEF_INDEL);
129 bcf_p1aux_t *bcf_p1_init(int n, uint8_t *ploidy)
133 ma = calloc(1, sizeof(bcf_p1aux_t));
135 ma->n = n; ma->M = 2 * n;
137 ma->ploidy = malloc(n);
138 memcpy(ma->ploidy, ploidy, n);
139 for (i = 0, ma->M = 0; i < n; ++i) ma->M += ploidy[i];
140 if (ma->M == 2 * n) {
145 ma->q2p = calloc(256, sizeof(double));
146 ma->pdg = calloc(3 * ma->n, sizeof(double));
147 ma->phi = calloc(ma->M + 1, sizeof(double));
148 ma->phi_indel = calloc(ma->M + 1, sizeof(double));
149 ma->phi1 = calloc(ma->M + 1, sizeof(double));
150 ma->phi2 = calloc(ma->M + 1, sizeof(double));
151 ma->z = calloc(ma->M + 1, sizeof(double));
152 ma->zswap = calloc(ma->M + 1, sizeof(double));
153 ma->z1 = calloc(ma->M + 1, sizeof(double)); // actually we do not need this large
154 ma->z2 = calloc(ma->M + 1, sizeof(double));
155 ma->afs = calloc(ma->M + 1, sizeof(double));
156 ma->afs1 = calloc(ma->M + 1, sizeof(double));
157 for (i = 0; i < 256; ++i)
158 ma->q2p[i] = pow(10., -i / 10.);
159 bcf_p1_init_prior(ma, MC_PTYPE_FULL, 1e-3); // the simplest prior
163 int bcf_p1_set_n1(bcf_p1aux_t *b, int n1)
165 if (n1 == 0 || n1 >= b->n) return -1;
166 if (b->M != b->n * 2) {
167 fprintf(stderr, "[%s] unable to set `n1' when there are haploid samples.\n", __func__);
174 void bcf_p1_destroy(bcf_p1aux_t *ma)
178 if (ma->hg && ma->n1 > 0) {
179 for (k = 0; k <= 2*ma->n1; ++k) free(ma->hg[k]);
182 free(ma->ploidy); free(ma->q2p); free(ma->pdg);
183 free(ma->phi); free(ma->phi_indel); free(ma->phi1); free(ma->phi2);
184 free(ma->z); free(ma->zswap); free(ma->z1); free(ma->z2);
185 free(ma->afs); free(ma->afs1);
190 static int cal_pdg(const bcf1_t *b, bcf_p1aux_t *ma)
194 p = alloca(b->n_alleles * sizeof(long));
195 memset(p, 0, sizeof(long) * b->n_alleles);
196 for (j = 0; j < ma->n; ++j) {
197 const uint8_t *pi = ma->PL + j * ma->PL_len;
198 double *pdg = ma->pdg + j * 3;
199 pdg[0] = ma->q2p[pi[2]]; pdg[1] = ma->q2p[pi[1]]; pdg[2] = ma->q2p[pi[0]];
200 for (i = 0; i < b->n_alleles; ++i)
201 p[i] += (int)pi[(i+1)*(i+2)/2-1];
203 for (i = 0; i < b->n_alleles; ++i) p[i] = p[i]<<4 | i;
204 for (i = 1; i < b->n_alleles; ++i) // insertion sort
205 for (j = i; j > 0 && p[j] < p[j-1]; --j)
206 tmp = p[j], p[j] = p[j-1], p[j-1] = tmp;
207 for (i = b->n_alleles - 1; i >= 0; --i)
208 if ((p[i]&0xf) == 0) break;
211 // f0 is the reference allele frequency
212 static double mc_freq_iter(double f0, const bcf_p1aux_t *ma, int beg, int end)
216 f3[0] = (1.-f0)*(1.-f0); f3[1] = 2.*f0*(1.-f0); f3[2] = f0*f0;
217 for (i = beg, f = 0.; i < end; ++i) {
219 pdg = ma->pdg + i * 3;
220 f += (pdg[1] * f3[1] + 2. * pdg[2] * f3[2])
221 / (pdg[0] * f3[0] + pdg[1] * f3[1] + pdg[2] * f3[2]);
223 f /= (end - beg) * 2.;
227 static double mc_gtfreq_iter(double g[3], const bcf_p1aux_t *ma, int beg, int end)
231 gg[0] = gg[1] = gg[2] = 0.;
232 for (i = beg; i < end; ++i) {
233 double *pdg, sum, tmp[3];
234 pdg = ma->pdg + i * 3;
235 tmp[0] = pdg[0] * g[0]; tmp[1] = pdg[1] * g[1]; tmp[2] = pdg[2] * g[2];
236 sum = (tmp[0] + tmp[1] + tmp[2]) * (end - beg);
237 gg[0] += tmp[0] / sum; gg[1] += tmp[1] / sum; gg[2] += tmp[2] / sum;
239 err = fabs(gg[0] - g[0]) > fabs(gg[1] - g[1])? fabs(gg[0] - g[0]) : fabs(gg[1] - g[1]);
240 err = err > fabs(gg[2] - g[2])? err : fabs(gg[2] - g[2]);
241 g[0] = gg[0]; g[1] = gg[1]; g[2] = gg[2];
245 int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int k)
248 double max, f3[3], *pdg = ma->pdg + k * 3;
249 int q, i, max_i, ploidy;
250 ploidy = ma->ploidy? ma->ploidy[k] : 2;
252 f3[0] = (1.-f0)*(1.-f0); f3[1] = 2.*f0*(1.-f0); f3[2] = f0*f0;
254 f3[0] = 1. - f0; f3[1] = 0; f3[2] = f0;
256 for (i = 0, sum = 0.; i < 3; ++i)
257 sum += (g[i] = pdg[i] * f3[i]);
258 for (i = 0, max = -1., max_i = 0; i < 3; ++i) {
260 if (g[i] > max) max = g[i], max_i = i;
263 if (max < 1e-308) max = 1e-308;
264 q = (int)(-4.343 * log(max) + .499);
271 static void mc_cal_y_core(bcf_p1aux_t *ma, int beg)
273 double *z[2], *tmp, *pdg;
274 int _j, last_min, last_max;
275 assert(beg == 0 || ma->M == ma->n*2);
279 memset(z[0], 0, sizeof(double) * (ma->M + 1));
280 memset(z[1], 0, sizeof(double) * (ma->M + 1));
282 last_min = last_max = 0;
284 if (ma->M == ma->n * 2) {
286 for (_j = beg; _j < ma->n; ++_j) {
287 int k, j = _j - beg, _min = last_min, _max = last_max, M0;
290 pdg = ma->pdg + _j * 3;
291 p[0] = pdg[0]; p[1] = 2. * pdg[1]; p[2] = pdg[2];
292 for (; _min < _max && z[0][_min] < TINY; ++_min) z[0][_min] = z[1][_min] = 0.;
293 for (; _max > _min && z[0][_max] < TINY; --_max) z[0][_max] = z[1][_max] = 0.;
295 if (_min == 0) k = 0, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k];
296 if (_min <= 1) k = 1, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k] + k*(M0-k+2) * p[1] * z[0][k-1];
297 for (k = _min < 2? 2 : _min; k <= _max; ++k)
298 z[1][k] = (M0-k+1)*(M0-k+2) * p[0] * z[0][k] + k*(M0-k+2) * p[1] * z[0][k-1] + k*(k-1)* p[2] * z[0][k-2];
299 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
300 ma->t += log(sum / (M * (M - 1.)));
301 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
302 if (_min >= 1) z[1][_min-1] = 0.;
303 if (_min >= 2) z[1][_min-2] = 0.;
304 if (j < ma->n - 1) z[1][_max+1] = z[1][_max+2] = 0.;
305 if (_j == ma->n1 - 1) { // set pop1; ma->n1==-1 when unset
307 memcpy(ma->z1, z[1], sizeof(double) * (ma->n1 * 2 + 1));
309 tmp = z[0]; z[0] = z[1]; z[1] = tmp;
310 last_min = _min; last_max = _max;
312 //for (_j = 0; _j < last_min; ++_j) z[0][_j] = 0.; // TODO: are these necessary?
313 //for (_j = last_max + 1; _j < ma->M; ++_j) z[0][_j] = 0.;
314 } else { // this block is very similar to the block above; these two might be merged in future
316 for (j = 0; j < ma->n; ++j) {
317 int k, M0, _min = last_min, _max = last_max;
319 pdg = ma->pdg + j * 3;
320 for (; _min < _max && z[0][_min] < TINY; ++_min) z[0][_min] = z[1][_min] = 0.;
321 for (; _max > _min && z[0][_max] < TINY; --_max) z[0][_max] = z[1][_max] = 0.;
324 if (ma->ploidy[j] == 1) {
325 p[0] = pdg[0]; p[1] = pdg[2];
327 if (_min == 0) k = 0, z[1][k] = (M0+1-k) * p[0] * z[0][k];
328 for (k = _min < 1? 1 : _min; k <= _max; ++k)
329 z[1][k] = (M0+1-k) * p[0] * z[0][k] + k * p[1] * z[0][k-1];
330 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
331 ma->t += log(sum / M);
332 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
333 if (_min >= 1) z[1][_min-1] = 0.;
334 if (j < ma->n - 1) z[1][_max+1] = 0.;
335 } else if (ma->ploidy[j] == 2) {
336 p[0] = pdg[0]; p[1] = 2 * pdg[1]; p[2] = pdg[2];
338 if (_min == 0) k = 0, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k];
339 if (_min <= 1) k = 1, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k] + k*(M0-k+2) * p[1] * z[0][k-1];
340 for (k = _min < 2? 2 : _min; k <= _max; ++k)
341 z[1][k] = (M0-k+1)*(M0-k+2) * p[0] * z[0][k] + k*(M0-k+2) * p[1] * z[0][k-1] + k*(k-1)* p[2] * z[0][k-2];
342 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
343 ma->t += log(sum / (M * (M - 1.)));
344 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
345 if (_min >= 1) z[1][_min-1] = 0.;
346 if (_min >= 2) z[1][_min-2] = 0.;
347 if (j < ma->n - 1) z[1][_max+1] = z[1][_max+2] = 0.;
349 tmp = z[0]; z[0] = z[1]; z[1] = tmp;
350 last_min = _min; last_max = _max;
353 if (z[0] != ma->z) memcpy(ma->z, z[0], sizeof(double) * (ma->M + 1));
356 static void mc_cal_y(bcf_p1aux_t *ma)
358 if (ma->n1 > 0 && ma->n1 < ma->n && ma->M == ma->n * 2) { // NB: ma->n1 is ineffective when there are haploid samples
361 memset(ma->z1, 0, sizeof(double) * (2 * ma->n1 + 1));
362 memset(ma->z2, 0, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
363 ma->t1 = ma->t2 = 0.;
364 mc_cal_y_core(ma, ma->n1);
366 memcpy(ma->z2, ma->z, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
367 mc_cal_y_core(ma, 0);
369 x = expl(ma->t - (ma->t1 + ma->t2));
370 for (k = 0; k <= ma->M; ++k) ma->z[k] *= x;
371 } else mc_cal_y_core(ma, 0);
374 #define CONTRAST_TINY 1e-30
376 extern double kf_gammaq(double s, double z); // incomplete gamma function for chi^2 test
378 static inline double chi2_test(int a, int b, int c, int d)
381 x = (double)(a+b) * (c+d) * (b+d) * (a+c);
382 if (x == 0.) return 1;
384 return kf_gammaq(.5, .5 * z * z * (a+b+c+d) / x);
387 // chi2=(a+b+c+d)(ad-bc)^2/[(a+b)(c+d)(a+c)(b+d)]
388 static inline double contrast2_aux(const bcf_p1aux_t *p1, double sum, int n1, int n2, int k1, int k2, double x[3])
390 double p = p1->phi[k1+k2] * p1->z1[k1] * p1->z2[k2] / sum * p1->hg[k1][k2];
391 if (p < CONTRAST_TINY) return -1;
392 if (.5*k1/n1 < .5*k2/n2) x[1] += p;
393 else if (.5*k1/n1 > .5*k2/n2) x[2] += p;
395 return p * chi2_test(k1, k2, (n1<<1) - k1, (n2<<1) - k2);
398 static double contrast2(bcf_p1aux_t *p1, double ret[3])
400 int k, k1, k2, k10, k20, n1, n2;
403 n1 = p1->n1; n2 = p1->n - p1->n1;
404 if (n1 <= 0 || n2 <= 0) return 0.;
405 if (p1->hg == 0) { // initialize the hypergeometric distribution
406 /* NB: the hg matrix may take a lot of memory when there are many samples. There is a way
407 to avoid precomputing this matrix, but it is slower and quite intricate. The following
408 computation in this block can be accelerated with a similar strategy, but perhaps this
409 is not a serious concern for now. */
410 double tmp = lgamma(2*(n1+n2)+1) - (lgamma(2*n1+1) + lgamma(2*n2+1));
411 p1->hg = calloc(2*n1+1, sizeof(void*));
412 for (k1 = 0; k1 <= 2*n1; ++k1) {
413 p1->hg[k1] = calloc(2*n2+1, sizeof(double));
414 for (k2 = 0; k2 <= 2*n2; ++k2)
415 p1->hg[k1][k2] = exp(lgamma(k1+k2+1) + lgamma(p1->M-k1-k2+1) - (lgamma(k1+1) + lgamma(k2+1) + lgamma(2*n1-k1+1) + lgamma(2*n2-k2+1) + tmp));
418 { // compute sum1 and sum2
419 long double suml = 0;
420 for (k = 0; k <= p1->M; ++k) suml += p1->phi[k] * p1->z[k];
423 { // get the mean k1 and k2
426 for (k = 0, max = 0, max_k = -1; k <= 2*n1; ++k) {
427 double x = p1->phi1[k] * p1->z1[k];
428 if (x > max) max = x, max_k = k;
431 for (k = 0, max = 0, max_k = -1; k <= 2*n2; ++k) {
432 double x = p1->phi2[k] * p1->z2[k];
433 if (x > max) max = x, max_k = k;
437 { // We can do the following with one nested loop, but that is an O(N^2) thing. The following code block is much faster for large N.
440 x[0] = x[1] = x[2] = 0;
441 for (k1 = k10; k1 >= 0; --k1) {
442 for (k2 = k20; k2 >= 0; --k2) {
443 if ((y = contrast2_aux(p1, sum, n1, n2, k1, k2, x)) < 0) break;
446 for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
447 if ((y = contrast2_aux(p1, sum, n1, n2, k1, k2, x)) < 0) break;
451 ret[0] = x[0]; ret[1] = x[1]; ret[2] = x[2];
452 x[0] = x[1] = x[2] = 0;
453 for (k1 = k10 + 1; k1 <= 2*n1; ++k1) {
454 for (k2 = k20; k2 >= 0; --k2) {
455 if ((y = contrast2_aux(p1, sum, n1, n2, k1, k2, x)) < 0) break;
458 for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
459 if ((y = contrast2_aux(p1, sum, n1, n2, k1, k2, x)) < 0) break;
463 ret[0] += x[0]; ret[1] += x[1]; ret[2] += x[2];
464 if (ret[0] + ret[1] + ret[2] < 0.99) { // in case of bad things happened
465 ret[0] = ret[1] = ret[2] = 0;
466 for (k1 = 0, z = 0.; k1 <= 2*n1; ++k1)
467 for (k2 = 0; k2 <= 2*n2; ++k2)
468 if ((y = contrast2_aux(p1, sum, n1, n2, k1, k2, ret)) >= 0) z += y;
469 if (ret[0] + ret[1] + ret[2] < 0.99) // It seems that this may be caused by floating point errors. I do not really understand why...
470 z = 1.0, ret[0] = ret[1] = ret[2] = 1./3;
476 static double mc_cal_afs(bcf_p1aux_t *ma, double *p_ref_folded, double *p_var_folded)
479 long double sum = 0., sum2;
480 double *phi = ma->is_indel? ma->phi_indel : ma->phi;
481 memset(ma->afs1, 0, sizeof(double) * (ma->M + 1));
484 for (k = 0, sum = 0.; k <= ma->M; ++k)
485 sum += (long double)phi[k] * ma->z[k];
486 for (k = 0; k <= ma->M; ++k) {
487 ma->afs1[k] = phi[k] * ma->z[k] / sum;
488 if (isnan(ma->afs1[k]) || isinf(ma->afs1[k])) return -1.;
490 // compute folded variant probability
491 for (k = 0, sum = 0.; k <= ma->M; ++k)
492 sum += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
493 for (k = 1, sum2 = 0.; k < ma->M; ++k)
494 sum2 += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
495 *p_var_folded = sum2 / sum;
496 *p_ref_folded = (phi[k] + phi[ma->M - k]) / 2. * (ma->z[ma->M] + ma->z[0]) / sum;
497 // the expected frequency
498 for (k = 0, sum = 0.; k <= ma->M; ++k) {
499 ma->afs[k] += ma->afs1[k];
500 sum += k * ma->afs1[k];
505 int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
508 long double sum = 0.;
509 ma->is_indel = bcf_is_indel(b);
512 for (i = 0; i < b->n_gi; ++i) {
513 if (b->gi[i].fmt == bcf_str2int("PL", 2)) {
514 ma->PL = (uint8_t*)b->gi[i].data;
515 ma->PL_len = b->gi[i].len;
519 if (b->n_alleles < 2) return -1; // FIXME: find a better solution
521 rst->rank0 = cal_pdg(b, ma);
522 rst->f_exp = mc_cal_afs(ma, &rst->p_ref_folded, &rst->p_var_folded);
523 rst->p_ref = ma->afs1[ma->M];
524 for (k = 0, sum = 0.; k < ma->M; ++k)
526 rst->p_var = (double)sum;
527 // calculate f_flat and f_em
528 for (k = 0, sum = 0.; k <= ma->M; ++k)
529 sum += (long double)ma->z[k];
531 for (k = 0; k <= ma->M; ++k) {
532 double p = ma->z[k] / sum;
533 rst->f_flat += k * p;
535 rst->f_flat /= ma->M;
537 double flast = rst->f_flat;
538 for (i = 0; i < MC_MAX_EM_ITER; ++i) {
539 rst->f_em = mc_freq_iter(flast, ma, 0, ma->n);
540 if (fabs(rst->f_em - flast) < MC_EM_EPS) break;
543 if (ma->n1 > 0 && ma->n1 < ma->n) {
544 for (k = 0; k < 2; ++k) {
546 for (i = 0; i < MC_MAX_EM_ITER; ++i) {
547 rst->f_em2[k] = k? mc_freq_iter(flast, ma, ma->n1, ma->n) : mc_freq_iter(flast, ma, 0, ma->n1);
548 if (fabs(rst->f_em2[k] - flast) < MC_EM_EPS) break;
549 flast = rst->f_em2[k];
555 rst->g[0] = (1. - rst->f_em) * (1. - rst->f_em);
556 rst->g[1] = 2. * rst->f_em * (1. - rst->f_em);
557 rst->g[2] = rst->f_em * rst->f_em;
558 for (i = 0; i < MC_MAX_EM_ITER; ++i)
559 if (mc_gtfreq_iter(rst->g, ma, 0, ma->n) < MC_EM_EPS) break;
561 { // estimate equal-tail credible interval (95% level)
564 for (i = 0, p = 0.; i < ma->M; ++i)
565 if (p + ma->afs1[i] > 0.025) break;
566 else p += ma->afs1[i];
568 for (i = ma->M-1, p = 0.; i >= 0; --i)
569 if (p + ma->afs1[i] > 0.025) break;
570 else p += ma->afs1[i];
572 rst->cil = (double)(ma->M - h) / ma->M; rst->cih = (double)(ma->M - l) / ma->M;
574 rst->cmp[0] = rst->cmp[1] = rst->cmp[2] = rst->p_chi2 = -1.0;
575 if (rst->p_var > 0.1) // skip contrast2() if the locus is a strong non-variant
576 rst->p_chi2 = contrast2(ma, rst->cmp);
580 void bcf_p1_dump_afs(bcf_p1aux_t *ma)
583 fprintf(stderr, "[afs]");
584 for (k = 0; k <= ma->M; ++k)
585 fprintf(stderr, " %d:%.3lf", k, ma->afs[ma->M - k]);
586 fprintf(stderr, "\n");
587 memset(ma->afs, 0, sizeof(double) * (ma->M + 1));