|
0
|
1 #include <math.h>
|
|
|
2 #include <stdlib.h>
|
|
|
3 #include <string.h>
|
|
|
4 #include <stdio.h>
|
|
|
5 #include <errno.h>
|
|
|
6 #include <assert.h>
|
|
|
7 #include "prob1.h"
|
|
|
8
|
|
|
9 #include "kseq.h"
|
|
|
10 KSTREAM_INIT(gzFile, gzread, 16384)
|
|
|
11
|
|
|
12 #define MC_MAX_EM_ITER 16
|
|
|
13 #define MC_EM_EPS 1e-5
|
|
|
14 #define MC_DEF_INDEL 0.15
|
|
|
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
|
|
|
33 };
|
|
|
34
|
|
|
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
|
|
|
43 double *lf; // log factorial
|
|
|
44 double t, t1, t2;
|
|
|
45 double *afs, *afs1; // afs: accumulative AFS; afs1: site posterior distribution
|
|
|
46 const uint8_t *PL; // point to PL
|
|
|
47 int PL_len;
|
|
|
48 };
|
|
|
49
|
|
|
50 void bcf_p1_indel_prior(bcf_p1aux_t *ma, double x)
|
|
|
51 {
|
|
|
52 int i;
|
|
|
53 for (i = 0; i < ma->M; ++i)
|
|
|
54 ma->phi_indel[i] = ma->phi[i] * x;
|
|
|
55 ma->phi_indel[ma->M] = 1. - ma->phi[ma->M] * x;
|
|
|
56 }
|
|
|
57
|
|
|
58 static void init_prior(int type, double theta, int M, double *phi)
|
|
|
59 {
|
|
|
60 int i;
|
|
|
61 if (type == MC_PTYPE_COND2) {
|
|
|
62 for (i = 0; i <= M; ++i)
|
|
|
63 phi[i] = 2. * (i + 1) / (M + 1) / (M + 2);
|
|
|
64 } else if (type == MC_PTYPE_FLAT) {
|
|
|
65 for (i = 0; i <= M; ++i)
|
|
|
66 phi[i] = 1. / (M + 1);
|
|
|
67 } else {
|
|
|
68 double sum;
|
|
|
69 for (i = 0, sum = 0.; i < M; ++i)
|
|
|
70 sum += (phi[i] = theta / (M - i));
|
|
|
71 phi[M] = 1. - sum;
|
|
|
72 }
|
|
|
73 }
|
|
|
74
|
|
|
75 void bcf_p1_init_prior(bcf_p1aux_t *ma, int type, double theta)
|
|
|
76 {
|
|
|
77 init_prior(type, theta, ma->M, ma->phi);
|
|
|
78 bcf_p1_indel_prior(ma, MC_DEF_INDEL);
|
|
|
79 }
|
|
|
80
|
|
|
81 void bcf_p1_init_subprior(bcf_p1aux_t *ma, int type, double theta)
|
|
|
82 {
|
|
|
83 if (ma->n1 <= 0 || ma->n1 >= ma->M) return;
|
|
|
84 init_prior(type, theta, 2*ma->n1, ma->phi1);
|
|
|
85 init_prior(type, theta, 2*(ma->n - ma->n1), ma->phi2);
|
|
|
86 }
|
|
|
87
|
|
|
88 int bcf_p1_read_prior(bcf_p1aux_t *ma, const char *fn)
|
|
|
89 {
|
|
|
90 gzFile fp;
|
|
|
91 kstring_t s;
|
|
|
92 kstream_t *ks;
|
|
|
93 long double sum;
|
|
|
94 int dret, k;
|
|
|
95 memset(&s, 0, sizeof(kstring_t));
|
|
|
96 fp = strcmp(fn, "-")? gzopen(fn, "r") : gzdopen(fileno(stdin), "r");
|
|
|
97 ks = ks_init(fp);
|
|
|
98 memset(ma->phi, 0, sizeof(double) * (ma->M + 1));
|
|
|
99 while (ks_getuntil(ks, '\n', &s, &dret) >= 0) {
|
|
|
100 if (strstr(s.s, "[afs] ") == s.s) {
|
|
|
101 char *p = s.s + 6;
|
|
|
102 for (k = 0; k <= ma->M; ++k) {
|
|
|
103 int x;
|
|
|
104 double y;
|
|
|
105 x = strtol(p, &p, 10);
|
|
|
106 if (x != k && (errno == EINVAL || errno == ERANGE)) return -1;
|
|
|
107 ++p;
|
|
|
108 y = strtod(p, &p);
|
|
|
109 if (y == 0. && (errno == EINVAL || errno == ERANGE)) return -1;
|
|
|
110 ma->phi[ma->M - k] += y;
|
|
|
111 }
|
|
|
112 }
|
|
|
113 }
|
|
|
114 ks_destroy(ks);
|
|
|
115 gzclose(fp);
|
|
|
116 free(s.s);
|
|
|
117 for (sum = 0., k = 0; k <= ma->M; ++k) sum += ma->phi[k];
|
|
|
118 fprintf(stderr, "[prior]");
|
|
|
119 for (k = 0; k <= ma->M; ++k) ma->phi[k] /= sum;
|
|
|
120 for (k = 0; k <= ma->M; ++k) fprintf(stderr, " %d:%.3lg", k, ma->phi[ma->M - k]);
|
|
|
121 fputc('\n', stderr);
|
|
|
122 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));
|
|
|
123 fprintf(stderr, "[%s] heterozygosity=%lf, ", __func__, (double)sum);
|
|
|
124 for (sum = 0., k = 1; k <= ma->M; ++k) sum += k * ma->phi[ma->M - k] / ma->M;
|
|
|
125 fprintf(stderr, "theta=%lf\n", (double)sum);
|
|
|
126 bcf_p1_indel_prior(ma, MC_DEF_INDEL);
|
|
|
127 return 0;
|
|
|
128 }
|
|
|
129
|
|
|
130 bcf_p1aux_t *bcf_p1_init(int n, uint8_t *ploidy)
|
|
|
131 {
|
|
|
132 bcf_p1aux_t *ma;
|
|
|
133 int i;
|
|
|
134 ma = calloc(1, sizeof(bcf_p1aux_t));
|
|
|
135 ma->n1 = -1;
|
|
|
136 ma->n = n; ma->M = 2 * n;
|
|
|
137 if (ploidy) {
|
|
|
138 ma->ploidy = malloc(n);
|
|
|
139 memcpy(ma->ploidy, ploidy, n);
|
|
|
140 for (i = 0, ma->M = 0; i < n; ++i) ma->M += ploidy[i];
|
|
|
141 if (ma->M == 2 * n) {
|
|
|
142 free(ma->ploidy);
|
|
|
143 ma->ploidy = 0;
|
|
|
144 }
|
|
|
145 }
|
|
|
146 ma->q2p = calloc(256, sizeof(double));
|
|
|
147 ma->pdg = calloc(3 * ma->n, sizeof(double));
|
|
|
148 ma->phi = calloc(ma->M + 1, sizeof(double));
|
|
|
149 ma->phi_indel = calloc(ma->M + 1, sizeof(double));
|
|
|
150 ma->phi1 = calloc(ma->M + 1, sizeof(double));
|
|
|
151 ma->phi2 = calloc(ma->M + 1, sizeof(double));
|
|
|
152 ma->z = calloc(ma->M + 1, sizeof(double));
|
|
|
153 ma->zswap = calloc(ma->M + 1, sizeof(double));
|
|
|
154 ma->z1 = calloc(ma->M + 1, sizeof(double)); // actually we do not need this large
|
|
|
155 ma->z2 = calloc(ma->M + 1, sizeof(double));
|
|
|
156 ma->afs = calloc(ma->M + 1, sizeof(double));
|
|
|
157 ma->afs1 = calloc(ma->M + 1, sizeof(double));
|
|
|
158 ma->lf = calloc(ma->M + 1, sizeof(double));
|
|
|
159 for (i = 0; i < 256; ++i)
|
|
|
160 ma->q2p[i] = pow(10., -i / 10.);
|
|
|
161 for (i = 0; i <= ma->M; ++i) ma->lf[i] = lgamma(i + 1);
|
|
|
162 bcf_p1_init_prior(ma, MC_PTYPE_FULL, 1e-3); // the simplest prior
|
|
|
163 return ma;
|
|
|
164 }
|
|
|
165
|
|
|
166 int bcf_p1_set_n1(bcf_p1aux_t *b, int n1)
|
|
|
167 {
|
|
|
168 if (n1 == 0 || n1 >= b->n) return -1;
|
|
|
169 if (b->M != b->n * 2) {
|
|
|
170 fprintf(stderr, "[%s] unable to set `n1' when there are haploid samples.\n", __func__);
|
|
|
171 return -1;
|
|
|
172 }
|
|
|
173 b->n1 = n1;
|
|
|
174 return 0;
|
|
|
175 }
|
|
|
176
|
|
|
177 void bcf_p1_destroy(bcf_p1aux_t *ma)
|
|
|
178 {
|
|
|
179 if (ma) {
|
|
|
180 int k;
|
|
|
181 free(ma->lf);
|
|
|
182 if (ma->hg && ma->n1 > 0) {
|
|
|
183 for (k = 0; k <= 2*ma->n1; ++k) free(ma->hg[k]);
|
|
|
184 free(ma->hg);
|
|
|
185 }
|
|
|
186 free(ma->ploidy); free(ma->q2p); free(ma->pdg);
|
|
|
187 free(ma->phi); free(ma->phi_indel); free(ma->phi1); free(ma->phi2);
|
|
|
188 free(ma->z); free(ma->zswap); free(ma->z1); free(ma->z2);
|
|
|
189 free(ma->afs); free(ma->afs1);
|
|
|
190 free(ma);
|
|
|
191 }
|
|
|
192 }
|
|
|
193
|
|
|
194 static int cal_pdg(const bcf1_t *b, bcf_p1aux_t *ma)
|
|
|
195 {
|
|
|
196 int i, j;
|
|
|
197 long *p, tmp;
|
|
|
198 p = alloca(b->n_alleles * sizeof(long));
|
|
|
199 memset(p, 0, sizeof(long) * b->n_alleles);
|
|
|
200 for (j = 0; j < ma->n; ++j) {
|
|
|
201 const uint8_t *pi = ma->PL + j * ma->PL_len;
|
|
|
202 double *pdg = ma->pdg + j * 3;
|
|
|
203 pdg[0] = ma->q2p[pi[2]]; pdg[1] = ma->q2p[pi[1]]; pdg[2] = ma->q2p[pi[0]];
|
|
|
204 for (i = 0; i < b->n_alleles; ++i)
|
|
|
205 p[i] += (int)pi[(i+1)*(i+2)/2-1];
|
|
|
206 }
|
|
|
207 for (i = 0; i < b->n_alleles; ++i) p[i] = p[i]<<4 | i;
|
|
|
208 for (i = 1; i < b->n_alleles; ++i) // insertion sort
|
|
|
209 for (j = i; j > 0 && p[j] < p[j-1]; --j)
|
|
|
210 tmp = p[j], p[j] = p[j-1], p[j-1] = tmp;
|
|
|
211 for (i = b->n_alleles - 1; i >= 0; --i)
|
|
|
212 if ((p[i]&0xf) == 0) break;
|
|
|
213 return i;
|
|
|
214 }
|
|
|
215
|
|
|
216 int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int k)
|
|
|
217 {
|
|
|
218 double sum, g[3];
|
|
|
219 double max, f3[3], *pdg = ma->pdg + k * 3;
|
|
|
220 int q, i, max_i, ploidy;
|
|
|
221 ploidy = ma->ploidy? ma->ploidy[k] : 2;
|
|
|
222 if (ploidy == 2) {
|
|
|
223 f3[0] = (1.-f0)*(1.-f0); f3[1] = 2.*f0*(1.-f0); f3[2] = f0*f0;
|
|
|
224 } else {
|
|
|
225 f3[0] = 1. - f0; f3[1] = 0; f3[2] = f0;
|
|
|
226 }
|
|
|
227 for (i = 0, sum = 0.; i < 3; ++i)
|
|
|
228 sum += (g[i] = pdg[i] * f3[i]);
|
|
|
229 for (i = 0, max = -1., max_i = 0; i < 3; ++i) {
|
|
|
230 g[i] /= sum;
|
|
|
231 if (g[i] > max) max = g[i], max_i = i;
|
|
|
232 }
|
|
|
233 max = 1. - max;
|
|
|
234 if (max < 1e-308) max = 1e-308;
|
|
|
235 q = (int)(-4.343 * log(max) + .499);
|
|
|
236 if (q > 99) q = 99;
|
|
|
237 return q<<2|max_i;
|
|
|
238 }
|
|
|
239
|
|
|
240 #define TINY 1e-20
|
|
|
241
|
|
|
242 static void mc_cal_y_core(bcf_p1aux_t *ma, int beg)
|
|
|
243 {
|
|
|
244 double *z[2], *tmp, *pdg;
|
|
|
245 int _j, last_min, last_max;
|
|
|
246 assert(beg == 0 || ma->M == ma->n*2);
|
|
|
247 z[0] = ma->z;
|
|
|
248 z[1] = ma->zswap;
|
|
|
249 pdg = ma->pdg;
|
|
|
250 memset(z[0], 0, sizeof(double) * (ma->M + 1));
|
|
|
251 memset(z[1], 0, sizeof(double) * (ma->M + 1));
|
|
|
252 z[0][0] = 1.;
|
|
|
253 last_min = last_max = 0;
|
|
|
254 ma->t = 0.;
|
|
|
255 if (ma->M == ma->n * 2) {
|
|
|
256 int M = 0;
|
|
|
257 for (_j = beg; _j < ma->n; ++_j) {
|
|
|
258 int k, j = _j - beg, _min = last_min, _max = last_max, M0;
|
|
|
259 double p[3], sum;
|
|
|
260 M0 = M; M += 2;
|
|
|
261 pdg = ma->pdg + _j * 3;
|
|
|
262 p[0] = pdg[0]; p[1] = 2. * pdg[1]; p[2] = pdg[2];
|
|
|
263 for (; _min < _max && z[0][_min] < TINY; ++_min) z[0][_min] = z[1][_min] = 0.;
|
|
|
264 for (; _max > _min && z[0][_max] < TINY; --_max) z[0][_max] = z[1][_max] = 0.;
|
|
|
265 _max += 2;
|
|
|
266 if (_min == 0) k = 0, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k];
|
|
|
267 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];
|
|
|
268 for (k = _min < 2? 2 : _min; k <= _max; ++k)
|
|
|
269 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];
|
|
|
270 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
|
|
|
271 ma->t += log(sum / (M * (M - 1.)));
|
|
|
272 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
|
|
|
273 if (_min >= 1) z[1][_min-1] = 0.;
|
|
|
274 if (_min >= 2) z[1][_min-2] = 0.;
|
|
|
275 if (j < ma->n - 1) z[1][_max+1] = z[1][_max+2] = 0.;
|
|
|
276 if (_j == ma->n1 - 1) { // set pop1; ma->n1==-1 when unset
|
|
|
277 ma->t1 = ma->t;
|
|
|
278 memcpy(ma->z1, z[1], sizeof(double) * (ma->n1 * 2 + 1));
|
|
|
279 }
|
|
|
280 tmp = z[0]; z[0] = z[1]; z[1] = tmp;
|
|
|
281 last_min = _min; last_max = _max;
|
|
|
282 }
|
|
|
283 //for (_j = 0; _j < last_min; ++_j) z[0][_j] = 0.; // TODO: are these necessary?
|
|
|
284 //for (_j = last_max + 1; _j < ma->M; ++_j) z[0][_j] = 0.;
|
|
|
285 } else { // this block is very similar to the block above; these two might be merged in future
|
|
|
286 int j, M = 0;
|
|
|
287 for (j = 0; j < ma->n; ++j) {
|
|
|
288 int k, M0, _min = last_min, _max = last_max;
|
|
|
289 double p[3], sum;
|
|
|
290 pdg = ma->pdg + j * 3;
|
|
|
291 for (; _min < _max && z[0][_min] < TINY; ++_min) z[0][_min] = z[1][_min] = 0.;
|
|
|
292 for (; _max > _min && z[0][_max] < TINY; --_max) z[0][_max] = z[1][_max] = 0.;
|
|
|
293 M0 = M;
|
|
|
294 M += ma->ploidy[j];
|
|
|
295 if (ma->ploidy[j] == 1) {
|
|
|
296 p[0] = pdg[0]; p[1] = pdg[2];
|
|
|
297 _max++;
|
|
|
298 if (_min == 0) k = 0, z[1][k] = (M0+1-k) * p[0] * z[0][k];
|
|
|
299 for (k = _min < 1? 1 : _min; k <= _max; ++k)
|
|
|
300 z[1][k] = (M0+1-k) * p[0] * z[0][k] + k * p[1] * z[0][k-1];
|
|
|
301 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
|
|
|
302 ma->t += log(sum / M);
|
|
|
303 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
|
|
|
304 if (_min >= 1) z[1][_min-1] = 0.;
|
|
|
305 if (j < ma->n - 1) z[1][_max+1] = 0.;
|
|
|
306 } else if (ma->ploidy[j] == 2) {
|
|
|
307 p[0] = pdg[0]; p[1] = 2 * pdg[1]; p[2] = pdg[2];
|
|
|
308 _max += 2;
|
|
|
309 if (_min == 0) k = 0, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k];
|
|
|
310 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];
|
|
|
311 for (k = _min < 2? 2 : _min; k <= _max; ++k)
|
|
|
312 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];
|
|
|
313 for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
|
|
|
314 ma->t += log(sum / (M * (M - 1.)));
|
|
|
315 for (k = _min; k <= _max; ++k) z[1][k] /= sum;
|
|
|
316 if (_min >= 1) z[1][_min-1] = 0.;
|
|
|
317 if (_min >= 2) z[1][_min-2] = 0.;
|
|
|
318 if (j < ma->n - 1) z[1][_max+1] = z[1][_max+2] = 0.;
|
|
|
319 }
|
|
|
320 tmp = z[0]; z[0] = z[1]; z[1] = tmp;
|
|
|
321 last_min = _min; last_max = _max;
|
|
|
322 }
|
|
|
323 }
|
|
|
324 if (z[0] != ma->z) memcpy(ma->z, z[0], sizeof(double) * (ma->M + 1));
|
|
|
325 }
|
|
|
326
|
|
|
327 static void mc_cal_y(bcf_p1aux_t *ma)
|
|
|
328 {
|
|
|
329 if (ma->n1 > 0 && ma->n1 < ma->n && ma->M == ma->n * 2) { // NB: ma->n1 is ineffective when there are haploid samples
|
|
|
330 int k;
|
|
|
331 long double x;
|
|
|
332 memset(ma->z1, 0, sizeof(double) * (2 * ma->n1 + 1));
|
|
|
333 memset(ma->z2, 0, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
|
|
|
334 ma->t1 = ma->t2 = 0.;
|
|
|
335 mc_cal_y_core(ma, ma->n1);
|
|
|
336 ma->t2 = ma->t;
|
|
|
337 memcpy(ma->z2, ma->z, sizeof(double) * (2 * (ma->n - ma->n1) + 1));
|
|
|
338 mc_cal_y_core(ma, 0);
|
|
|
339 // rescale z
|
|
|
340 x = expl(ma->t - (ma->t1 + ma->t2));
|
|
|
341 for (k = 0; k <= ma->M; ++k) ma->z[k] *= x;
|
|
|
342 } else mc_cal_y_core(ma, 0);
|
|
|
343 }
|
|
|
344
|
|
|
345 #define CONTRAST_TINY 1e-30
|
|
|
346
|
|
|
347 extern double kf_gammaq(double s, double z); // incomplete gamma function for chi^2 test
|
|
|
348
|
|
|
349 static inline double chi2_test(int a, int b, int c, int d)
|
|
|
350 {
|
|
|
351 double x, z;
|
|
|
352 x = (double)(a+b) * (c+d) * (b+d) * (a+c);
|
|
|
353 if (x == 0.) return 1;
|
|
|
354 z = a * d - b * c;
|
|
|
355 return kf_gammaq(.5, .5 * z * z * (a+b+c+d) / x);
|
|
|
356 }
|
|
|
357
|
|
|
358 // chi2=(a+b+c+d)(ad-bc)^2/[(a+b)(c+d)(a+c)(b+d)]
|
|
|
359 static inline double contrast2_aux(const bcf_p1aux_t *p1, double sum, int k1, int k2, double x[3])
|
|
|
360 {
|
|
|
361 double p = p1->phi[k1+k2] * p1->z1[k1] * p1->z2[k2] / sum * p1->hg[k1][k2];
|
|
|
362 int n1 = p1->n1, n2 = p1->n - p1->n1;
|
|
|
363 if (p < CONTRAST_TINY) return -1;
|
|
|
364 if (.5*k1/n1 < .5*k2/n2) x[1] += p;
|
|
|
365 else if (.5*k1/n1 > .5*k2/n2) x[2] += p;
|
|
|
366 else x[0] += p;
|
|
|
367 return p * chi2_test(k1, k2, (n1<<1) - k1, (n2<<1) - k2);
|
|
|
368 }
|
|
|
369
|
|
|
370 static double contrast2(bcf_p1aux_t *p1, double ret[3])
|
|
|
371 {
|
|
|
372 int k, k1, k2, k10, k20, n1, n2;
|
|
|
373 double sum;
|
|
|
374 // get n1 and n2
|
|
|
375 n1 = p1->n1; n2 = p1->n - p1->n1;
|
|
|
376 if (n1 <= 0 || n2 <= 0) return 0.;
|
|
|
377 if (p1->hg == 0) { // initialize the hypergeometric distribution
|
|
|
378 /* NB: the hg matrix may take a lot of memory when there are many samples. There is a way
|
|
|
379 to avoid precomputing this matrix, but it is slower and quite intricate. The following
|
|
|
380 computation in this block can be accelerated with a similar strategy, but perhaps this
|
|
|
381 is not a serious concern for now. */
|
|
|
382 double tmp = lgamma(2*(n1+n2)+1) - (lgamma(2*n1+1) + lgamma(2*n2+1));
|
|
|
383 p1->hg = calloc(2*n1+1, sizeof(void*));
|
|
|
384 for (k1 = 0; k1 <= 2*n1; ++k1) {
|
|
|
385 p1->hg[k1] = calloc(2*n2+1, sizeof(double));
|
|
|
386 for (k2 = 0; k2 <= 2*n2; ++k2)
|
|
|
387 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));
|
|
|
388 }
|
|
|
389 }
|
|
|
390 { // compute
|
|
|
391 long double suml = 0;
|
|
|
392 for (k = 0; k <= p1->M; ++k) suml += p1->phi[k] * p1->z[k];
|
|
|
393 sum = suml;
|
|
|
394 }
|
|
|
395 { // get the max k1 and k2
|
|
|
396 double max;
|
|
|
397 int max_k;
|
|
|
398 for (k = 0, max = 0, max_k = -1; k <= 2*n1; ++k) {
|
|
|
399 double x = p1->phi1[k] * p1->z1[k];
|
|
|
400 if (x > max) max = x, max_k = k;
|
|
|
401 }
|
|
|
402 k10 = max_k;
|
|
|
403 for (k = 0, max = 0, max_k = -1; k <= 2*n2; ++k) {
|
|
|
404 double x = p1->phi2[k] * p1->z2[k];
|
|
|
405 if (x > max) max = x, max_k = k;
|
|
|
406 }
|
|
|
407 k20 = max_k;
|
|
|
408 }
|
|
|
409 { // 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.
|
|
|
410 double x[3], y;
|
|
|
411 long double z = 0., L[2];
|
|
|
412 x[0] = x[1] = x[2] = 0; L[0] = L[1] = 0;
|
|
|
413 for (k1 = k10; k1 >= 0; --k1) {
|
|
|
414 for (k2 = k20; k2 >= 0; --k2) {
|
|
|
415 if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
|
|
|
416 else z += y;
|
|
|
417 }
|
|
|
418 for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
|
|
|
419 if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
|
|
|
420 else z += y;
|
|
|
421 }
|
|
|
422 }
|
|
|
423 ret[0] = x[0]; ret[1] = x[1]; ret[2] = x[2];
|
|
|
424 x[0] = x[1] = x[2] = 0;
|
|
|
425 for (k1 = k10 + 1; k1 <= 2*n1; ++k1) {
|
|
|
426 for (k2 = k20; k2 >= 0; --k2) {
|
|
|
427 if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
|
|
|
428 else z += y;
|
|
|
429 }
|
|
|
430 for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
|
|
|
431 if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
|
|
|
432 else z += y;
|
|
|
433 }
|
|
|
434 }
|
|
|
435 ret[0] += x[0]; ret[1] += x[1]; ret[2] += x[2];
|
|
|
436 if (ret[0] + ret[1] + ret[2] < 0.95) { // in case of bad things happened
|
|
|
437 ret[0] = ret[1] = ret[2] = 0; L[0] = L[1] = 0;
|
|
|
438 for (k1 = 0, z = 0.; k1 <= 2*n1; ++k1)
|
|
|
439 for (k2 = 0; k2 <= 2*n2; ++k2)
|
|
|
440 if ((y = contrast2_aux(p1, sum, k1, k2, ret)) >= 0) z += y;
|
|
|
441 if (ret[0] + ret[1] + ret[2] < 0.95) // It seems that this may be caused by floating point errors. I do not really understand why...
|
|
|
442 z = 1.0, ret[0] = ret[1] = ret[2] = 1./3;
|
|
|
443 }
|
|
|
444 return (double)z;
|
|
|
445 }
|
|
|
446 }
|
|
|
447
|
|
|
448 static double mc_cal_afs(bcf_p1aux_t *ma, double *p_ref_folded, double *p_var_folded)
|
|
|
449 {
|
|
|
450 int k;
|
|
|
451 long double sum = 0., sum2;
|
|
|
452 double *phi = ma->is_indel? ma->phi_indel : ma->phi;
|
|
|
453 memset(ma->afs1, 0, sizeof(double) * (ma->M + 1));
|
|
|
454 mc_cal_y(ma);
|
|
|
455 // compute AFS
|
|
|
456 for (k = 0, sum = 0.; k <= ma->M; ++k)
|
|
|
457 sum += (long double)phi[k] * ma->z[k];
|
|
|
458 for (k = 0; k <= ma->M; ++k) {
|
|
|
459 ma->afs1[k] = phi[k] * ma->z[k] / sum;
|
|
|
460 if (isnan(ma->afs1[k]) || isinf(ma->afs1[k])) return -1.;
|
|
|
461 }
|
|
|
462 // compute folded variant probability
|
|
|
463 for (k = 0, sum = 0.; k <= ma->M; ++k)
|
|
|
464 sum += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
|
|
|
465 for (k = 1, sum2 = 0.; k < ma->M; ++k)
|
|
|
466 sum2 += (long double)(phi[k] + phi[ma->M - k]) / 2. * ma->z[k];
|
|
|
467 *p_var_folded = sum2 / sum;
|
|
|
468 *p_ref_folded = (phi[k] + phi[ma->M - k]) / 2. * (ma->z[ma->M] + ma->z[0]) / sum;
|
|
|
469 // the expected frequency
|
|
|
470 for (k = 0, sum = 0.; k <= ma->M; ++k) {
|
|
|
471 ma->afs[k] += ma->afs1[k];
|
|
|
472 sum += k * ma->afs1[k];
|
|
|
473 }
|
|
|
474 return sum / ma->M;
|
|
|
475 }
|
|
|
476
|
|
|
477 int bcf_p1_cal(const bcf1_t *b, int do_contrast, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
|
|
|
478 {
|
|
|
479 int i, k;
|
|
|
480 long double sum = 0.;
|
|
|
481 ma->is_indel = bcf_is_indel(b);
|
|
|
482 rst->perm_rank = -1;
|
|
|
483 // set PL and PL_len
|
|
|
484 for (i = 0; i < b->n_gi; ++i) {
|
|
|
485 if (b->gi[i].fmt == bcf_str2int("PL", 2)) {
|
|
|
486 ma->PL = (uint8_t*)b->gi[i].data;
|
|
|
487 ma->PL_len = b->gi[i].len;
|
|
|
488 break;
|
|
|
489 }
|
|
|
490 }
|
|
|
491 if (i == b->n_gi) return -1; // no PL
|
|
|
492 if (b->n_alleles < 2) return -1; // FIXME: find a better solution
|
|
|
493 //
|
|
|
494 rst->rank0 = cal_pdg(b, ma);
|
|
|
495 rst->f_exp = mc_cal_afs(ma, &rst->p_ref_folded, &rst->p_var_folded);
|
|
|
496 rst->p_ref = ma->afs1[ma->M];
|
|
|
497 for (k = 0, sum = 0.; k < ma->M; ++k)
|
|
|
498 sum += ma->afs1[k];
|
|
|
499 rst->p_var = (double)sum;
|
|
|
500 // calculate f_flat and f_em
|
|
|
501 for (k = 0, sum = 0.; k <= ma->M; ++k)
|
|
|
502 sum += (long double)ma->z[k];
|
|
|
503 rst->f_flat = 0.;
|
|
|
504 for (k = 0; k <= ma->M; ++k) {
|
|
|
505 double p = ma->z[k] / sum;
|
|
|
506 rst->f_flat += k * p;
|
|
|
507 }
|
|
|
508 rst->f_flat /= ma->M;
|
|
|
509 { // estimate equal-tail credible interval (95% level)
|
|
|
510 int l, h;
|
|
|
511 double p;
|
|
|
512 for (i = 0, p = 0.; i < ma->M; ++i)
|
|
|
513 if (p + ma->afs1[i] > 0.025) break;
|
|
|
514 else p += ma->afs1[i];
|
|
|
515 l = i;
|
|
|
516 for (i = ma->M-1, p = 0.; i >= 0; --i)
|
|
|
517 if (p + ma->afs1[i] > 0.025) break;
|
|
|
518 else p += ma->afs1[i];
|
|
|
519 h = i;
|
|
|
520 rst->cil = (double)(ma->M - h) / ma->M; rst->cih = (double)(ma->M - l) / ma->M;
|
|
|
521 }
|
|
|
522 rst->cmp[0] = rst->cmp[1] = rst->cmp[2] = rst->p_chi2 = -1.0;
|
|
|
523 if (do_contrast && rst->p_var > 0.5) // skip contrast2() if the locus is a strong non-variant
|
|
|
524 rst->p_chi2 = contrast2(ma, rst->cmp);
|
|
|
525 return 0;
|
|
|
526 }
|
|
|
527
|
|
|
528 void bcf_p1_dump_afs(bcf_p1aux_t *ma)
|
|
|
529 {
|
|
|
530 int k;
|
|
|
531 fprintf(stderr, "[afs]");
|
|
|
532 for (k = 0; k <= ma->M; ++k)
|
|
|
533 fprintf(stderr, " %d:%.3lf", k, ma->afs[ma->M - k]);
|
|
|
534 fprintf(stderr, "\n");
|
|
|
535 memset(ma->afs, 0, sizeof(double) * (ma->M + 1));
|
|
|
536 }
|
