// qual:6, strand:1, base:4
int errmod_cal(const errmod_t *em, int n, int m, uint16_t *bases, float *q)
{
call_aux_t aux;
int i, j, k, w[32];
if (m > m) return -1;
memset(q, 0, m * m * sizeof(float));
if (n == 0) return 0;
// calculate aux.esum and aux.fsum
if (n > 255) { // then sample 255 bases
ks_shuffle(uint16_t, n, bases);
n = 255;
}
ks_introsort(uint16_t, n, bases);
memset(w, 0, 32 * sizeof(int));
memset(&aux, 0, sizeof(call_aux_t));
for (j = n - 1; j >= 0; --j) { // calculate esum and fsum
uint16_t b = bases[j];
int q = b>>5 < 4? 4 : b>>5;
if (q > 63) q = 63;
k = b&0x1f;
aux.fsum[k&0xf] += em->coef->fk[w[k]];
aux.bsum[k&0xf] += em->coef->fk[w[k]] * em->coef->beta[q<<16|n<<8|aux.c[k&0xf]];
++aux.c[k&0xf];
++w[k];
}
// generate likelihood
for (j = 0; j != m; ++j) {
float tmp1, tmp3;
int tmp2, bar_e;
// homozygous
for (k = 0, tmp1 = tmp3 = 0.0, tmp2 = 0; k != m; ++k) {
if (k == j) continue;
tmp1 += aux.bsum[k]; tmp2 += aux.c[k]; tmp3 += aux.fsum[k];
}
if (tmp2) {
bar_e = (int)(tmp1 / tmp3 + 0.499);
if (bar_e > 63) bar_e = 63;
q[j*m+j] = tmp1;
}
// heterozygous
for (k = j + 1; k < m; ++k) {
int cjk = aux.c[j] + aux.c[k];
for (i = 0, tmp2 = 0, tmp1 = tmp3 = 0.0; i < m; ++i) {
if (i == j || i == k) continue;
tmp1 += aux.bsum[i]; tmp2 += aux.c[i]; tmp3 += aux.fsum[i];
}
if (tmp2) {
bar_e = (int)(tmp1 / tmp3 + 0.499);
if (bar_e > 63) bar_e = 63;
q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]] + tmp1;
} else q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]]; // all the bases are either j or k
}
for (k = 0; k != m; ++k) if (q[j*m+k] < 0.0) q[j*m+k] = 0.0;
}
return 0;
}
//
// em: error model to fit to data
// m: number of alleles across all samples
// n: number of bases observed in sample
// bases[i]: bases observed in pileup [6 bit quality|1 bit strand|4 bit base]
// q[i*m+j]: (Output) phred-scaled likelihood of each genotype (i,j)
int errmod_cal(const errmod_t *em, int n, int m, uint16_t *bases, float *q)
{
// Aux
// aux.c is total count of each base observed (ignoring strand)
call_aux_t aux;
// Loop variables
int i, j, k;
// The total count of each base observed per strand
int w[32];
/* zero out q */
memset(q, 0, m * m * sizeof(float));
if (n == 0) return 0;
// calculate aux.esum and aux.fsum
if (n > 255) { // then sample 255 bases
ks_shuffle(uint16_t, n, bases);
n = 255;
}
ks_introsort(uint16_t, n, bases);
/* zero out w and aux */
memset(w, 0, 32 * sizeof(int));
memset(&aux, 0, sizeof(call_aux_t));
for (j = n - 1; j >= 0; --j) { // calculate esum and fsum
uint16_t b = bases[j];
/* extract quality and cap at 63 */
int qual = b>>5 < 4? 4 : b>>5;
if (qual > 63) qual = 63;
/* extract base ORed with strand */
int basestrand = b&0x1f;
/* extract base */
int base = b&0xf;
aux.fsum[base] += em->coef->fk[w[basestrand]];
aux.bsum[base] += em->coef->fk[w[basestrand]] * em->coef->beta[qual<<16|n<<8|aux.c[base]];
++aux.c[base];
++w[basestrand];
}
// generate likelihood
for (j = 0; j < m; ++j) {
float tmp1, tmp3;
int tmp2;
// homozygous
for (k = 0, tmp1 = tmp3 = 0.0, tmp2 = 0; k < m; ++k) {
if (k == j) continue;
tmp1 += aux.bsum[k]; tmp2 += aux.c[k]; tmp3 += aux.fsum[k];
}
if (tmp2) {
q[j*m+j] = tmp1;
}
// heterozygous
for (k = j + 1; k < m; ++k) {
int cjk = aux.c[j] + aux.c[k];
for (i = 0, tmp2 = 0, tmp1 = tmp3 = 0.0; i < m; ++i) {
if (i == j || i == k) continue;
tmp1 += aux.bsum[i]; tmp2 += aux.c[i]; tmp3 += aux.fsum[i];
}
if (tmp2) {
q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]] + tmp1;
} else q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]]; // all the bases are either j or k
}
/* clamp to greater than 0 */
for (k = 0; k < m; ++k) if (q[j*m+k] < 0.0) q[j*m+k] = 0.0;
}
return 0;
}
// qual:6, strand:1, base:4
int errmod_cal(const errmod_t *em, int n, int m, uint16_t *bases, float *q)
{
// fprintf(stderr,"n=%d\n",n);
if (n == 0)
return 0; //no data
//reset results array;
memset(q, 0, m * m * sizeof(float));
//set "counts" array
int w[32];
memset(w, 0, 32 * sizeof(int));
for(int i=0;0&&i<32;i++)
fprintf(stderr,"pre w[%d]=%d\n",i,w[i]);
call_aux_t aux;
memset(&aux, 0, sizeof(call_aux_t));
//sample 255 if depth >255
if ((n > 255)) { // then sample 255 bases //THIS MAKES
ks_shuffle(uint16_t, n, bases);
n = 255;
}
ks_introsort(uint16_t, n, bases);
for (int j = n - 1; j >= 0; --j) {
uint16_t b = bases[j];
// fprintf(stderr,"j=%d q=%d strand=%d base=%d\n",j,b>>5,0x1&(b>>4),b&0xf);
// fprintf(stderr,"j=%d q=%d\tc=%c\t",j,(b>>5) + 33,(b>>5) + 33);
//cap quality at [4,63]
int q = b>>5 < 4? 4 : b>>5;
if (q > 63)
q = 63;
int k = b&0x1f;
aux.bsum[k&0xf] += em->coef->fk[w[k]] * em->coef->beta[q<<16|n<<8|aux.c[k&0xf]];
++aux.c[k&0xf];
++w[k];
for(int i=0;0&&i<32;i++)
fprintf(stderr,"w[%d]=%d\n",i,w[i]);
}
if(0){
//floating point inprecision compared with samtools binary output. But is correct
//the genotype like p(data|A1=g1,A2=g2) = p(data|A1=g2,A2=g1)
for (int g1 = 0; g1 <5; ++g1) {//allele1=0,1,2,3,4
for (int g2 = g1; g2<5; ++g2) {//allele2=0,1,2,3,4
if(g1!=g2){ // A1!=A2 - heterozygoues
int cjk = aux.c[g1] + aux.c[g2];//total depth for allele g1+allele g2
//binomial when ignoring non A1/A2 alleles: Bin (n,k,p);n=cjk=#g1+#g2 , k= aux.c[g2] = #g2 ; p=0.5. returns log
q[g1*5+g2] = -4.343 * em->coef->lhet[cjk<<8|aux.c[g2]];
}
for (int k = 0; k <5; ++k){
if(k!=g1 && k!=g2) //if a read has a non A1/A2 alleles it is an error. add the log of the prob of these reads
q[g1*5+g2] += aux.bsum[k];
}
//mirror
if(g1!=g2)
q[g2*5+g1] = q[g1*5+g2];
if (q[g1*5+g2] < 0.0)
q[g1*5+g2] = 0.0;
}
}
return 0;
}
// generate likelihood THIS WORKS PERFECTLY june 4 ande
for (int g1 = 0; g1 <5; ++g1) {//j=0,1,2,3,4
for (int g2 = g1; g2<5; ++g2) {//j=0,1,2,3,4
if(g1==g2){
for (int k = 0; k <5; ++k){
if(k!=g1)
q[g1*5+g2] += aux.bsum[k];
}
}
else{
int other=0;
float tmp1=0;
for (int k = 0; k < 5; ++k)
if (k != g1 && k != g2) {
tmp1 += aux.bsum[k];
other = 1;
}
int cjk = aux.c[g1] + aux.c[g2];
if (other)
q[g1*5+g2] = q[g2*5+g1] = -4.343 * em->coef->lhet[cjk<<8|aux.c[g2]] + tmp1;
else
q[g1*5+g2] = q[g2*5+g1] = -4.343 * em->coef->lhet[cjk<<8|aux.c[g2]]; // all the bases are either j or k
}
if (q[g1*5+g2] < 0.0)
q[g1*5+g2] = 0.0;
}
}
return 0;
//old original almost
for (int j = 0; j != m; ++j) {//j=0,1,2,3,4
// homozygous
for (int k = 0; k != m; ++k){//only updates if k!=j and aux.c[k]!=0
fprintf(stderr,"\t-> j=%d aux.c[%d]=%d aux.bsum[%d]=%f res=%d ",j,k,aux.c[k],k,aux.bsum[k],j*m+j);
if (k != j && aux.c[k]) {
fprintf(stderr,"USING \n");
q[j*m+j] += aux.bsum[k];
}else
fprintf(stderr,"skipping\n");
}
// heterozygous
for (int k = j + 1; k < m; ++k) {//k=1,...,4
float tmp1=0.0;
int isHe=0;
for (int i = 0; i < m; ++i)
if (i != j && i != k) {
tmp1 += aux.bsum[i];
isHe += aux.c[i];
}
int cjk = aux.c[j] + aux.c[k];
fprintf(stderr,"j=%d k=%d RES=%d\n",j,k,j*m+k);
if(1) {
if (isHe)
q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]] + tmp1;
else
q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]]; // all the bases are either j or k
}else{
q[j*m+k] = q[k*m+j] = -4.343 * em->coef->lhet[cjk<<8|aux.c[k]];
if(isHe)
q[j*m+k] = q[k*m+j] = q[k*m+j]+tmp1;
}
}
//set to zero if negative, shoulnd't happen
for (int k = 0; k != m; ++k)
if (q[j*m+k] < 0.0)
q[j*m+k] = 0.0;
}
// exit(0);
return 0;
}