X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bcftools%2Fprob1.c;h=83bd8e2881f2f84688ce287b9a2ebe5e785cdace;hb=9f118264ea012adc21a46d7c03eaad4f9ce7d4d4;hp=4804e6e24c3c6787f2ca3a18fb6e83687f379ed4;hpb=f528e7da717a1a2a6ab2672c7c55f37d315452f3;p=samtools.git

diff --git a/bcftools/prob1.c b/bcftools/prob1.c
index 4804e6e..83bd8e2 100644
--- a/bcftools/prob1.c
+++ b/bcftools/prob1.c
@@ -10,7 +10,7 @@
 KSTREAM_INIT(gzFile, gzread, 16384)
 
 #define MC_MAX_EM_ITER 16
-#define MC_EM_EPS 1e-4
+#define MC_EM_EPS 1e-5
 #define MC_DEF_INDEL 0.15
 
 unsigned char seq_nt4_table[256] = {
@@ -39,6 +39,8 @@ struct __bcf_p1aux_t {
 	double *phi, *phi_indel;
 	double *z, *zswap; // aux for afs
 	double *z1, *z2, *phi1, *phi2; // only calculated when n1 is set
+	double **hg; // hypergeometric distribution
+	double *lf; // log factorial
 	double t, t1, t2;
 	double *afs, *afs1; // afs: accumulative AFS; afs1: site posterior distribution
 	const uint8_t *PL; // point to PL
@@ -153,8 +155,10 @@ bcf_p1aux_t *bcf_p1_init(int n, uint8_t *ploidy)
 	ma->z2 = calloc(ma->M + 1, sizeof(double));
 	ma->afs = calloc(ma->M + 1, sizeof(double));
 	ma->afs1 = calloc(ma->M + 1, sizeof(double));
+	ma->lf = calloc(ma->M + 1, sizeof(double));
 	for (i = 0; i < 256; ++i)
 		ma->q2p[i] = pow(10., -i / 10.);
+	for (i = 0; i <= ma->M; ++i) ma->lf[i] = lgamma(i + 1);
 	bcf_p1_init_prior(ma, MC_PTYPE_FULL, 1e-3); // the simplest prior
 	return ma;
 }
@@ -173,6 +177,12 @@ int bcf_p1_set_n1(bcf_p1aux_t *b, int n1)
 void bcf_p1_destroy(bcf_p1aux_t *ma)
 {
 	if (ma) {
+		int k;
+		free(ma->lf);
+		if (ma->hg && ma->n1 > 0) {
+			for (k = 0; k <= 2*ma->n1; ++k) free(ma->hg[k]);
+			free(ma->hg);
+		}
 		free(ma->ploidy); free(ma->q2p); free(ma->pdg);
 		free(ma->phi); free(ma->phi_indel); free(ma->phi1); free(ma->phi2);
 		free(ma->z); free(ma->zswap); free(ma->z1); free(ma->z2);
@@ -184,38 +194,49 @@ void bcf_p1_destroy(bcf_p1aux_t *ma)
 static int cal_pdg(const bcf1_t *b, bcf_p1aux_t *ma)
 {
 	int i, j;
-	long *p, tmp;
-	p = alloca(b->n_alleles * sizeof(long));
-	memset(p, 0, sizeof(long) * b->n_alleles);
+    int n = (b->n_alleles+1)*b->n_alleles/2;
+    double *lk = alloca(n * sizeof(long));
+    memset(lk, 0, sizeof(double) * n);
 	for (j = 0; j < ma->n; ++j) {
 		const uint8_t *pi = ma->PL + j * ma->PL_len;
 		double *pdg = ma->pdg + j * 3;
 		pdg[0] = ma->q2p[pi[2]]; pdg[1] = ma->q2p[pi[1]]; pdg[2] = ma->q2p[pi[0]];
-		for (i = 0; i < b->n_alleles; ++i)
-			p[i] += (int)pi[(i+1)*(i+2)/2-1];
-	}
-	for (i = 0; i < b->n_alleles; ++i) p[i] = p[i]<<4 | i;
-	for (i = 1; i < b->n_alleles; ++i) // insertion sort
-		for (j = i; j > 0 && p[j] < p[j-1]; --j)
-			tmp = p[j], p[j] = p[j-1], p[j-1] = tmp;
-	for (i = b->n_alleles - 1; i >= 0; --i)
-		if ((p[i]&0xf) == 0) break;
-	return i;
-}
-// f0 is the reference allele frequency
-static double mc_freq_iter(double f0, const bcf_p1aux_t *ma)
-{
-	double f, f3[3];
-	int i;
-	f3[0] = (1.-f0)*(1.-f0); f3[1] = 2.*f0*(1.-f0); f3[2] = f0*f0;
-	for (i = 0, f = 0.; i < ma->n; ++i) {
-		double *pdg;
-		pdg = ma->pdg + i * 3;
-		f += (pdg[1] * f3[1] + 2. * pdg[2] * f3[2])
-			/ (pdg[0] * f3[0] + pdg[1] * f3[1] + pdg[2] * f3[2]);
-	}
-	f /= ma->n * 2.;
-	return f;
+        for (i=0; i<n; i++) lk[i] += pi[i];
+    }
+
+    double norm=lk[0]; 
+    for (i=1; i<n; i++) if (lk[i]<norm) norm=lk[i];
+    #if DBG
+    for (i=0,j=0; i<b->n_alleles; i++)
+    {
+        int k; for (k=0; k<=i; k++) printf("%.0f\t", lk[j++]);
+        printf("\n");
+    }
+    #endif
+    for (i=0; i<n; i++) lk[i] = pow(10,-0.1*(lk[i]-norm));
+
+    // Find out the most likely alleles. In contrast to the original version,
+    //  ALT alleles may not be printed when they are more likely than REF but
+    //  significantly less likely than the most likely ALT. The only criterion
+    //  is the LK ratio now. To obtain behaviour similar to the original one,
+    //  use the pref variant below.
+    double pmax=0; //,pref=0;
+    n = ma->is_indel ? b->n_alleles : b->n_alleles-1;
+    for (i=0; i<n; i++)
+    {
+        double pr=0;
+        int k=i*(i+1)/2;
+        for (j=0; j<=i; j++) { pr+=lk[k]; k++; }
+        for (j=i+1; j<b->n_alleles; j++) { k=j*(j+1)/2+i; pr+=lk[k]; }
+        #if DBG
+        printf("%d\t%e\n", i,pr);
+        #endif
+        if (pmax<pr) pmax=pr;
+        // if (i==0) pref=pr;
+        // if (pr<pref && pr/pmax < 1e-4) break;
+        if (pr/pmax < 1e-4) break;   // Assuming the alleles are sorted by the lk
+    }
+	return i-1;
 }
 
 int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int k)
@@ -231,7 +252,7 @@ int bcf_p1_call_gt(const bcf_p1aux_t *ma, double f0, int k)
 	}
 	for (i = 0, sum = 0.; i < 3; ++i)
 		sum += (g[i] = pdg[i] * f3[i]);
-	for (i = 0, max = -1., max_i = 0; i <= ploidy; ++i) {
+	for (i = 0, max = -1., max_i = 0; i < 3; ++i) {
 		g[i] /= sum;
 		if (g[i] > max) max = g[i], max_i = i;
 	}
@@ -258,24 +279,22 @@ static void mc_cal_y_core(bcf_p1aux_t *ma, int beg)
 	last_min = last_max = 0;
 	ma->t = 0.;
 	if (ma->M == ma->n * 2) {
+		int M = 0;
 		for (_j = beg; _j < ma->n; ++_j) {
-			int k, j = _j - beg, _min = last_min, _max = last_max;
+			int k, j = _j - beg, _min = last_min, _max = last_max, M0;
 			double p[3], sum;
+			M0 = M; M += 2;
 			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.;
 			_max += 2;
-			if (_min == 0) 
-				k = 0, z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k];
-			if (_min <= 1)
-				k = 1, z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k] + k*(2*j+2-k) * p[1] * z[0][k-1];
+			if (_min == 0) k = 0, z[1][k] = (M0-k+1) * (M0-k+2) * p[0] * z[0][k];
+			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];
 			for (k = _min < 2? 2 : _min; k <= _max; ++k)
-				z[1][k] = (2*j+2-k)*(2*j-k+1) * p[0] * z[0][k]
-					+ k*(2*j+2-k) * p[1] * z[0][k-1]
-					+ k*(k-1)* p[2] * z[0][k-2];
+				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];
 			for (k = _min, sum = 0.; k <= _max; ++k) sum += z[1][k];
-			ma->t += log(sum / ((2. * j + 2) * (2. * j + 1)));
+			ma->t += log(sum / (M * (M - 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.;
@@ -287,6 +306,8 @@ static void mc_cal_y_core(bcf_p1aux_t *ma, int beg)
 			tmp = z[0]; z[0] = z[1]; z[1] = tmp;
 			last_min = _min; last_max = _max;
 		}
+		//for (_j = 0; _j < last_min; ++_j) z[0][_j] = 0.; // TODO: are these necessary?
+		//for (_j = last_max + 1; _j < ma->M; ++_j) z[0][_j] = 0.;
 	} else { // this block is very similar to the block above; these two might be merged in future
 		int j, M = 0;
 		for (j = 0; j < ma->n; ++j) {
@@ -347,26 +368,107 @@ static void mc_cal_y(bcf_p1aux_t *ma)
 	} 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
+#define CONTRAST_TINY 1e-30
+
+extern double kf_gammaq(double s, double z); // incomplete gamma function for chi^2 test
+
+static inline double chi2_test(int a, int b, int c, int d)
+{
+	double x, z;
+	x = (double)(a+b) * (c+d) * (b+d) * (a+c);
+	if (x == 0.) return 1;
+	z = a * d - b * c;
+	return kf_gammaq(.5, .5 * z * z * (a+b+c+d) / x);
+}
+
+// chi2=(a+b+c+d)(ad-bc)^2/[(a+b)(c+d)(a+c)(b+d)]
+static inline double contrast2_aux(const bcf_p1aux_t *p1, double sum, int k1, int k2, double x[3])
+{
+	double p = p1->phi[k1+k2] * p1->z1[k1] * p1->z2[k2] / sum * p1->hg[k1][k2];
+	int n1 = p1->n1, n2 = p1->n - p1->n1;
+	if (p < CONTRAST_TINY) return -1;
+	if (.5*k1/n1 < .5*k2/n2) x[1] += p;
+	else if (.5*k1/n1 > .5*k2/n2) x[2] += p;
+	else x[0] += p;
+	return p * chi2_test(k1, k2, (n1<<1) - k1, (n2<<1) - k2);
+}
+
+static double contrast2(bcf_p1aux_t *p1, double ret[3])
 {
-	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;
+	int k, k1, k2, k10, k20, n1, n2;
+	double sum;
+	// get n1 and n2
+	n1 = p1->n1; n2 = p1->n - p1->n1;
+	if (n1 <= 0 || n2 <= 0) return 0.;
+	if (p1->hg == 0) { // initialize the hypergeometric distribution
+		/* NB: the hg matrix may take a lot of memory when there are many samples. There is a way
+		   to avoid precomputing this matrix, but it is slower and quite intricate. The following
+		   computation in this block can be accelerated with a similar strategy, but perhaps this
+		   is not a serious concern for now. */
+		double tmp = lgamma(2*(n1+n2)+1) - (lgamma(2*n1+1) + lgamma(2*n2+1));
+		p1->hg = calloc(2*n1+1, sizeof(void*));
+		for (k1 = 0; k1 <= 2*n1; ++k1) {
+			p1->hg[k1] = calloc(2*n2+1, sizeof(double));
+			for (k2 = 0; k2 <= 2*n2; ++k2)
+				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));
+		}
+	}
+	{ // compute
+		long double suml = 0;
+		for (k = 0; k <= p1->M; ++k) suml += p1->phi[k] * p1->z[k];
+		sum = suml;
+	}
+	{ // get the max k1 and k2
+		double max;
+		int max_k;
+		for (k = 0, max = 0, max_k = -1; k <= 2*n1; ++k) {
+			double x = p1->phi1[k] * p1->z1[k];
+			if (x > max) max = x, max_k = k;
+		}
+		k10 = max_k;
+		for (k = 0, max = 0, max_k = -1; k <= 2*n2; ++k) {
+			double x = p1->phi2[k] * p1->z2[k];
+			if (x > max) max = x, max_k = k;
+		}
+		k20 = max_k;
+	}
+	{ // 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.
+		double x[3], y;
+		long double z = 0., L[2];
+		x[0] = x[1] = x[2] = 0; L[0] = L[1] = 0;
+		for (k1 = k10; k1 >= 0; --k1) {
+			for (k2 = k20; k2 >= 0; --k2) {
+				if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
+				else z += y;
+			}
+			for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
+				if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
+				else z += y;
+			}
+		}
+		ret[0] = x[0]; ret[1] = x[1]; ret[2] = x[2];
+		x[0] = x[1] = x[2] = 0;
+		for (k1 = k10 + 1; k1 <= 2*n1; ++k1) {
+			for (k2 = k20; k2 >= 0; --k2) {
+				if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
+				else z += y;
+			}
+			for (k2 = k20 + 1; k2 <= 2*n2; ++k2) {
+				if ((y = contrast2_aux(p1, sum, k1, k2, x)) < 0) break;
+				else z += y;
+			}
+		}
+		ret[0] += x[0]; ret[1] += x[1]; ret[2] += x[2];
+		if (ret[0] + ret[1] + ret[2] < 0.95) { // in case of bad things happened
+			ret[0] = ret[1] = ret[2] = 0; L[0] = L[1] = 0;
+			for (k1 = 0, z = 0.; k1 <= 2*n1; ++k1)
+				for (k2 = 0; k2 <= 2*n2; ++k2)
+					if ((y = contrast2_aux(p1, sum, k1, k2, ret)) >= 0) z += y;
+			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...
+				z = 1.0, ret[0] = ret[1] = ret[2] = 1./3;
+		}
+		return (double)z;
 	}
-	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, double *p_ref_folded, double *p_var_folded)
@@ -398,11 +500,12 @@ static double mc_cal_afs(bcf_p1aux_t *ma, double *p_ref_folded, double *p_var_fo
 	return sum / ma->M;
 }
 
-int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
+int bcf_p1_cal(const bcf1_t *b, int do_contrast, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
 {
 	int i, k;
 	long double sum = 0.;
 	ma->is_indel = bcf_is_indel(b);
+	rst->perm_rank = -1;
 	// set PL and PL_len
 	for (i = 0; i < b->n_gi; ++i) {
 		if (b->gi[i].fmt == bcf_str2int("PL", 2)) {
@@ -411,6 +514,7 @@ int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
 			break;
 		}
 	}
+	if (i == b->n_gi) return -1; // no PL
 	if (b->n_alleles < 2) return -1; // FIXME: find a better solution
 	// 
 	rst->rank0 = cal_pdg(b, ma);
@@ -419,6 +523,13 @@ int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
 	for (k = 0, sum = 0.; k < ma->M; ++k)
 		sum += ma->afs1[k];
 	rst->p_var = (double)sum;
+	{ // compute the allele count
+		double max = -1;
+		rst->ac = -1;
+		for (k = 0; k <= ma->M; ++k)
+			if (max < ma->z[k]) max = ma->z[k], rst->ac = k;
+		rst->ac = ma->M - rst->ac;
+	}
 	// calculate f_flat and f_em
 	for (k = 0, sum = 0.; k <= ma->M; ++k)
 		sum += (long double)ma->z[k];
@@ -428,29 +539,33 @@ int bcf_p1_cal(const bcf1_t *b, bcf_p1aux_t *ma, bcf_p1rst_t *rst)
 		rst->f_flat += k * p;
 	}
 	rst->f_flat /= ma->M;
-	{ // calculate f_em
-		double flast = rst->f_flat;
-		for (i = 0; i < MC_MAX_EM_ITER; ++i) {
-			rst->f_em = mc_freq_iter(flast, ma);
-			if (fabs(rst->f_em - flast) < MC_EM_EPS) break;
-			flast = rst->f_em;
-		}
-	}
 	{ // estimate equal-tail credible interval (95% level)
 		int l, h;
 		double p;
-		for (i = 0, p = 0.; i < ma->M; ++i)
+		for (i = 0, p = 0.; i <= ma->M; ++i)
 			if (p + ma->afs1[i] > 0.025) break;
 			else p += ma->afs1[i];
 		l = i;
-		for (i = ma->M-1, p = 0.; i >= 0; --i)
+		for (i = ma->M, p = 0.; i >= 0; --i)
 			if (p + ma->afs1[i] > 0.025) break;
 			else p += ma->afs1[i];
 		h = i;
 		rst->cil = (double)(ma->M - h) / ma->M; rst->cih = (double)(ma->M - l) / ma->M;
 	}
-	rst->g[0] = rst->g[1] = rst->g[2] = -1.;
-	contrast(ma, rst->pc);
+	if (ma->n1 > 0) { // compute LRT
+		double max0, max1, max2;
+		for (k = 0, max0 = -1; k <= ma->M; ++k)
+			if (max0 < ma->z[k]) max0 = ma->z[k];
+		for (k = 0, max1 = -1; k <= ma->n1 * 2; ++k)
+			if (max1 < ma->z1[k]) max1 = ma->z1[k];
+		for (k = 0, max2 = -1; k <= ma->M - ma->n1 * 2; ++k)
+			if (max2 < ma->z2[k]) max2 = ma->z2[k];
+		rst->lrt = log(max1 * max2 / max0);
+		rst->lrt = rst->lrt < 0? 1 : kf_gammaq(.5, rst->lrt);
+	} else rst->lrt = -1.0;
+	rst->cmp[0] = rst->cmp[1] = rst->cmp[2] = rst->p_chi2 = -1.0;
+	if (do_contrast && rst->p_var > 0.5) // skip contrast2() if the locus is a strong non-variant
+		rst->p_chi2 = contrast2(ma, rst->cmp);
 	return 0;
 }