void setFreq(perChr &pc){
pc.keeps=new char*[pc.nSites]; // array nSites x nInd 0 if missing info
pc.keepInd = new int[pc.nSites];
for(int s=0;s<pc.nSites;s++) {
pc.keeps[s] = new char[pc.nInd];
int nKeep =0;
for(int i=0;i<pc.nInd;i++){
double mmin=std::min(pc.gl[i][s*3],std::min(pc.gl[i][s*3+1],pc.gl[i][s*3+2]));
double mmax=std::max(pc.gl[i][s*3],std::max(pc.gl[i][s*3+1],pc.gl[i][s*3+2]));
if(fabs(mmax-mmin)<misTol)
pc.keeps[s][i] =0;
else{
pc.keeps[s][i] =1;
nKeep++;
}
}
pc.keepInd[s] = nKeep;
}
double tmp[3*pc.nInd];
pc.freq = new double[ pc.nSites];
pc.qerf = new double [pc.nSites];
for(int s=0;s<pc.nSites;s++){
for(int i=0;i<pc.nInd;i++)
for(int o=0;o<3;o++)
tmp[i*3+o] = pc.gl[i][s*3+o];
double af = emFrequency(tmp,pc.nInd,MAF_ITER,MAF_START,pc.keeps[s],pc.keepInd[s]);
assert(af<=1 && af>=0);
pc.freq[s] = log(af);
pc.qerf[s] = log(1-af);
}
}
void frequency::likeFreq(funkyPars *pars){//method=1: bfgs_known ;method=2 em;method=4 bfgs_unknown
//here only the likelihoods for the three genotypes are used.
double **loglike = NULL;
if(inputIsBeagle==1)
loglike= pars->likes;
else
loglike=angsd::get3likesRescale(pars);
assert(loglike!=NULL);
//the pml frequencies
double *pml=NULL;
double *pEM =NULL;
double *pmlun =NULL;
double *pEMun =NULL;
double *pmlSNP=NULL;
double *pEMSNP =NULL;
double *pmlunSNP =NULL;
double *pEMunSNP =NULL;
double *returnFreq = new double[pars->numSites]; //return value not to be deleted
double *lrt_snp = new double[pars->numSites]; //return value not to be deleted
// fprintf(stderr,"method:=%d\n",method);
if(doMaf &1){
pml = new double[pars->numSites]; // numeric optimisation
if(doSNP)
pmlSNP = new double[pars->numSites]; // numeric optimisation
}if(doMaf &2){
pEM =new double[pars->numSites]; //em algorithm
for(int i=0;i<pars->numSites;i++)
pEM[i] = 0.0;
if(doSNP)
pEMSNP =new double[pars->numSites]; //em algorithm
}if(doMaf &4){
pmlun = new double[pars->numSites]; // numeric optimisation
if(doSNP)
pmlunSNP = new double[pars->numSites]; // numeric optimisation
}if(doMaf &8){
pEMun = new double[pars->numSites]; // EM optimisation
if(doSNP)
pEMunSNP = new double[pars->numSites]; // EM optimisation
}
// number of individuals with data
int *keepInd = pars->keepSites;
int keepList[pars->nInd];
//loop though all sites and check if we have data.
//fprintf(stderr,"keepSites[0] %d\n",pars->keepSites[0]);
for(int s=0;s<pars->numSites;s++) {
if(keepInd[s]==0)//if we dont have any information
continue;
keepInd[s]=0;//
for(int i=0 ; i<pars->nInd ;i++) {//DRAGON CHECK THIS
//fprintf(stderr,"size %d\nind %d\t loglike:%f\t%f\t%f\n",s,i,loglike[s][i*3+0],loglike[s][i*3+1],loglike[s][i*3+2]);
keepList[i]=1;
if(loglike[s][i*3+0]+loglike[s][i*3+1]+loglike[s][i*3+2]>-0.0001){
// fprintf(stderr,"size %d\nind %d\t loglike:%f\t%f\t%f\n",s,i,loglike[s][i*3+0],loglike[s][i*3+1],loglike[s][i*3+2]);
keepList[i]=0;
}
else{
keepInd[s]++;
}
}
if(doMaf &1) {
pml[s]= likeFixedMinor_bfgs(loglike[s],pars->nInd);
if(doSNP &1)
pmlSNP[s] = 2*likeFixedMinor(0.0,loglike[s],pars->nInd)-2*likeFixedMinor(pml[s],loglike[s],pars->nInd);
}
if( doMaf &2) {
if(pars->phat!=NULL)
pEM[s]=emFrequency(loglike[s],pars->nInd,emIter,pars->phat[s],keepList,keepInd[s]);
else
pEM[s]=emFrequency(loglike[s],pars->nInd,emIter,EM_START,keepList,keepInd[s]);
if(doSNP)
pEMSNP[s] = 2*likeFixedMinor(0.0,loglike[s],pars->nInd)-2*likeFixedMinor(pEM[s],loglike[s],pars->nInd);
}
if( doMaf &4) {
pmlun[s]= likeNoFixedMinor_bfgs(pars->likes[s],pars->nInd,pars->major[s]);
if(doSNP)
pmlunSNP[s]= 2*likeNoFixedMinor(0.0,pars->likes[s],pars->nInd,pars->major[s])-2*likeNoFixedMinor(pmlun[s],pars->likes[s],pars->nInd,pars->major[s]);
}
if( doMaf &8 ){
if(pars->phat!=NULL)
pEMun[s]=emFrequencyNoFixed(pars->likes[s],pars->nInd,emIter,pars->phat[s],keepList,keepInd[s],pars->major[s],pars->minor[s]);
else
pEMun[s]=emFrequencyNoFixed(pars->likes[s],pars->nInd,emIter,EM_START,keepList,keepInd[s],pars->major[s],pars->minor[s]);
if(doSNP)
pEMunSNP[s] = 2*likeNoFixedMinor(0.0,pars->likes[s],pars->nInd,pars->major[s])-2*likeNoFixedMinor(pEMun[s],pars->likes[s],pars->nInd,pars->major[s]);
}
}
pars->results->freq->pml=pml;
pars->results->freq->pEM=pEM;
pars->results->freq->pmlun=pmlun;
pars->results->freq->pEMun=pEMun;
pars->results->freq->pmlSNP=pmlSNP;
pars->results->freq->pEMSNP=pEMSNP;
pars->results->freq->pmlunSNP=pmlunSNP;
pars->results->freq->pEMunSNP=pEMunSNP;
if(inputIsBeagle!=1){
for(int i=0;i<pars->numSites;i++)
delete [] loglike[i];
delete [] loglike;
}
//anders: this is less stupid but slower
//this is abit stupid, but in the case where multiple MAFS has been selected, we use the max(method)
//thorfinn: this is very stupid
for(int s=0;s<pars->numSites;s++){
if(doMaf &8)
returnFreq[s]=pEMun[s];
else if(doMaf &4)
returnFreq[s]=pmlun[s];
else if(doMaf &2 )
returnFreq[s]=pEM[s];
else if(doMaf &1 )
returnFreq[s]=pml[s];
}
//thorfinn april 16 2012
for(int s=0;s<pars->numSites;s++){
if((doMaf &8) && doSNP)
lrt_snp[s]=pEMunSNP[s];
else if((doMaf &4) && doSNP)
lrt_snp[s]=pmlunSNP[s];
else if((doMaf &2) && doSNP )
lrt_snp[s]=pEMSNP[s];
else if((doMaf &1) && doSNP)
lrt_snp[s]=pmlSNP[s];
}
pars->results->asso->freq=returnFreq;
pars->results->asso->lrt_snp = lrt_snp;//thorfinn add 16april 2012
}
