Compress binary packages with xz.

[samtools.git] / bam2bcf.c

diff --git a/bam2bcf.c b/bam2bcf.c
index 088635c0b04a549b762d1a3b28efad82f4bb5c80..dec3305340f689a100d29dc4f9b5998e7692c6d4 100644 (file)
--- a/bam2bcf.c
+++ b/bam2bcf.c
@@ -11,6 +11,7 @@ extern        void ks_introsort_uint32_t(size_t n, uint32_t a[]);
 #define CALL_ETA 0.03f
 #define CALL_MAX 256
 #define CALL_DEFTHETA 0.83f
+#define DEF_MAPQ 20
 
 #define CAP_DIST 25
 
@@ -23,6 +24,8 @@ bcf_callaux_t *bcf_call_init(double theta, int min_baseQ)
        bca->openQ = 40; bca->extQ = 20; bca->tandemQ = 100;
        bca->min_baseQ = min_baseQ;
        bca->e = errmod_init(1. - theta);
+       bca->min_frac = 0.002;
+       bca->min_support = 1;
        return bca;
 }
 
@@ -36,6 +39,7 @@ void bcf_call_destroy(bcf_callaux_t *bca)
  * negative if we are looking at an indel. */
 int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t *bca, bcf_callret1_t *r)
 {
+    static int *var_pos = NULL, nvar_pos = 0;
        int i, n, ref4, is_indel, ori_depth = 0;
        memset(r, 0, sizeof(bcf_callret1_t));
        if (ref_base >= 0) {
@@ -61,7 +65,8 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t
                seqQ = is_indel? (p->aux>>8&0xff) : 99;
                if (q < bca->min_baseQ) continue;
                if (q > seqQ) q = seqQ;
-               mapQ = p->b->core.qual < bca->capQ? p->b->core.qual : bca->capQ;
+               mapQ = p->b->core.qual < 255? p->b->core.qual : DEF_MAPQ; // special case for mapQ==255
+               mapQ = mapQ < bca->capQ? mapQ : bca->capQ;
                if (q > mapQ) q = mapQ;
                if (q > 63) q = 63;
                if (q < 4) q = 4;
@@ -75,7 +80,7 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t
                }
                bca->bases[n++] = q<<5 | (int)bam1_strand(p->b)<<4 | b;
                // collect annotations
-               r->qsum[b] += q;
+               if (b < 4) r->qsum[b] += q;
                ++r->anno[0<<2|is_diff<<1|bam1_strand(p->b)];
                min_dist = p->b->core.l_qseq - 1 - p->qpos;
                if (min_dist > p->qpos) min_dist = p->qpos;
@@ -90,9 +95,92 @@ int bcf_call_glfgen(int _n, const bam_pileup1_t *pl, int ref_base, bcf_callaux_t
        r->depth = n; r->ori_depth = ori_depth;
        // glfgen
        errmod_cal(bca->e, n, 5, bca->bases, r->p);
+
+    // Calculate the Variant Distance Bias (make it optional?)
+    if ( nvar_pos < _n ) {
+        nvar_pos = _n;
+        var_pos = realloc(var_pos,sizeof(int)*nvar_pos);
+    }
+    int alt_dp=0, read_len=0;
+    for (i=0; i<_n; i++) {
+        const bam_pileup1_t *p = pl + i;
+        if ( bam1_seqi(bam1_seq(p->b),p->qpos) == ref_base ) 
+            continue;
+
+        var_pos[alt_dp] = p->qpos;
+        if ( (bam1_cigar(p->b)[0]&BAM_CIGAR_MASK)==4 )
+            var_pos[alt_dp] -= bam1_cigar(p->b)[0]>>BAM_CIGAR_SHIFT;
+
+        alt_dp++;
+        read_len += p->b->core.l_qseq;
+    }
+    float mvd=0;
+    int j;
+    n=0;
+    for (i=0; i<alt_dp; i++) {
+        for (j=0; j<i; j++) {
+            mvd += abs(var_pos[i] - var_pos[j]);
+            n++;
+        }
+    }
+    r->mvd[0] = n ? mvd/n : 0;
+    r->mvd[1] = alt_dp;
+    r->mvd[2] = alt_dp ? read_len/alt_dp : 0;
+
        return r->depth;
 }
 
+
+void calc_vdb(int n, const bcf_callret1_t *calls, bcf_call_t *call)
+{
+    // Variant distance bias. Samples merged by means of DP-weighted average.
+
+    float weight=0, tot_prob=0;
+
+    int i;
+    for (i=0; i<n; i++)
+    {
+        int mvd      = calls[i].mvd[0];
+        int dp       = calls[i].mvd[1];
+        int read_len = calls[i].mvd[2];
+
+        if ( dp<2 ) continue;
+
+        float prob = 0;
+        if ( dp==2 )
+        {
+            // Exact formula
+            prob = (mvd==0) ? 1.0/read_len : (read_len-mvd)*2.0/read_len/read_len;
+        }
+        else if ( dp==3 )
+        {
+            // Sin, quite accurate approximation
+            float mu = read_len/2.9;
+            prob = mvd>2*mu ? 0 : sin(mvd*3.14/2/mu) / (4*mu/3.14);
+        }
+        else
+        {
+            // Scaled gaussian curve, crude approximation, but behaves well. Using fixed depth for bigger depths.
+            if ( dp>5 )
+                dp = 5;
+            float sigma2 = (read_len/1.9/(dp+1)) * (read_len/1.9/(dp+1));
+            float norm   = 1.125*sqrt(2*3.14*sigma2);
+            float mu     = read_len/2.9;
+            if ( mvd < mu )
+                prob = exp(-(mvd-mu)*(mvd-mu)/2/sigma2)/norm;
+            else
+                prob = exp(-(mvd-mu)*(mvd-mu)/3.125/sigma2)/norm;
+        }
+
+        //fprintf(stderr,"dp=%d mvd=%d read_len=%d -> prob=%f\n", dp,mvd,read_len,prob);
+        tot_prob += prob*dp;
+        weight += dp;
+    }
+    tot_prob = weight ? tot_prob/weight : 1; 
+    //fprintf(stderr,"prob=%f\n", tot_prob);
+    call->vdb = tot_prob;
+}
+
 int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/, bcf_call_t *call)
 {
        int ref4, i, j, qsum[4];
@@ -140,8 +228,8 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/,
                x = call->n_alleles * (call->n_alleles + 1) / 2;
                // get the possible genotypes
                for (i = z = 0; i < call->n_alleles; ++i)
-                       for (j = i; j < call->n_alleles; ++j)
-                               g[z++] = call->a[i] * 5 + call->a[j];
+                       for (j = 0; j <= i; ++j)
+                               g[z++] = call->a[j] * 5 + call->a[i];
                for (i = 0; i < n; ++i) {
                        uint8_t *PL = call->PL + x * i;
                        const bcf_callret1_t *r = calls + i;
@@ -166,6 +254,9 @@ int bcf_call_combine(int n, const bcf_callret1_t *calls, int ref_base /*4-bit*/,
                call->ori_depth += calls[i].ori_depth;
                for (j = 0; j < 16; ++j) call->anno[j] += calls[i].anno[j];
        }
+
+    calc_vdb(n, calls, call);
+
        return 0;
 }
 
@@ -219,6 +310,10 @@ int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bc
                if (i) kputc(',', &s);
                kputw(bc->anno[i], &s);
        }
+    if ( bc->vdb!=1 )
+    {
+        ksprintf(&s, ";VDB=%.4f", bc->vdb);
+    }
        kputc('\0', &s);
        // FMT
        kputs("PL", &s);
@@ -232,7 +327,7 @@ int bcf_call2bcf(int tid, int pos, bcf_call_t *bc, bcf1_t *b, bcf_callret1_t *bc
        memcpy(b->gi[0].data, bc->PL, b->gi[0].len * bc->n);
        if (bcr) {
                uint16_t *dp = (uint16_t*)b->gi[1].data;
-               uint8_t *sp = is_SP? b->gi[2].data : 0;
+               int32_t *sp = is_SP? b->gi[2].data : 0;
                for (i = 0; i < bc->n; ++i) {
                        bcf_callret1_t *p = bcr + i;
                        dp[i] = p->depth < 0xffff? p->depth : 0xffff;