void gcta::blup_snp_dosage()
{
check_autosome();
if(_mu.empty()) calcu_mu();
int i=0, j=0, k=0, col_num=_varcmp_Py.cols();
// Substract each element by 2p
for(i=0; i<_keep.size(); i++){
for(j=0; j<_include.size(); j++) _geno_dose[i][_include[j]]-=_mu[_include[j]];
}
// Calcuate A matrix
cout<<"Calculating the BLUP solution to SNP effects using imputed dosage scores ... "<<endl;
vector<double> var_SNP(_include.size()); // variance of each SNP, 2pq
eigenMatrix b_SNP=eigenMatrix::Zero(_include.size(), col_num); // variance of each SNP, 2pq
for(j=0; j<_include.size(); j++){
for(i=0; i<_keep.size(); i++) var_SNP[j]+=_geno_dose[i][_include[j]]*_geno_dose[i][_include[j]];
var_SNP[j]/=(double)(_keep.size()-1);
if(fabs(var_SNP[j])<1.0e-50) var_SNP[j]=0.0;
else var_SNP[j]=1.0/var_SNP[j];
}
for(k=0; k<_include.size(); k++){
for(i=0; i<_keep.size(); i++){
for(j=0; j<col_num; j++) b_SNP(k,j)+=_geno_dose[i][_include[k]]*_varcmp_Py(i,j);
}
for(j=0; j<col_num; j++) b_SNP(k,j)=b_SNP(k,j)*var_SNP[k]/(double)_include.size();
}
output_blup_snp(b_SNP);
}
// blue estimate of SNP effect
void gcta::blup_snp_geno()
{
check_autosome();
if(_mu.empty()) calcu_mu();
int i=0, j=0, k=0, col_num=_varcmp_Py.cols();
double x=0.0, fcount=0.0;
// Calcuate A matrix
cout<<"Calculating the BLUP solution to SNP effects ..."<<endl;
vector<double> var_SNP(_include.size());
eigenMatrix b_SNP=eigenMatrix::Zero(_include.size(), col_num); // variance of each SNP, 2pq
for(j=0; j<_include.size(); j++){
var_SNP[j]=_mu[_include[j]]*(1.0-0.5*_mu[_include[j]]);
if(fabs(var_SNP[j])<1.0e-50) var_SNP[j]=0.0;
else var_SNP[j]=1.0/var_SNP[j];
}
for(k=0; k<_include.size(); k++){
fcount=0.0;
for(i=0; i<_keep.size(); i++){
if(!_snp_1[_include[k]][i] || _snp_2[_include[k]][i]){
if(_allele1[_include[k]]==_ref_A[_include[k]]) x=_snp_1[_include[k]][i]+_snp_2[_include[k]][i];
else x=2.0-(_snp_1[_include[k]][i]+_snp_2[_include[k]][i]);
x=(x-_mu[_include[k]]);
for(j=0; j<col_num; j++) b_SNP(k,j)+=x*_varcmp_Py(i,j);
fcount+=1.0;
}
}
for(j=0; j<col_num; j++) b_SNP(k,j)=(b_SNP(k,j)*var_SNP[k]/fcount)*((double)_keep.size()/(double)_include.size());
}
output_blup_snp(b_SNP);
}
void gcta::save_XMat(bool miss_with_mu)
{
if(miss_with_mu && _mu.empty()) calcu_mu();
// Save matrix X
string X_zFile=_out+".xmat.gz";
gzofstream zoutf;
zoutf.open( X_zFile.c_str() );
if(!zoutf.is_open()) throw("Error: can not open the file ["+X_zFile+"] to write.");
cout<<"Saving the recoded genotype matrix to the file ["+X_zFile+"]."<<endl;
int i=0, j=0;
zoutf<<"FID IID ";
for(j=0; j<_include.size(); j++) zoutf<<_snp_name[_include[j]]<<" ";
zoutf<<endl;
zoutf<<"Reference Allele ";
for(j=0; j<_include.size(); j++) zoutf<<_ref_A[_include[j]]<<" ";
zoutf<<endl;
for(i=0; i<_keep.size(); i++){
zoutf<<_fid[_keep[i]]<<' '<<_pid[_keep[i]]<<' ';
if(_dosage_flag){
for(j=0; j<_include.size(); j++){
if(_geno_dose[_keep[i]][_include[j]]<1e5){
if(_allele1[_include[j]]==_ref_A[_include[j]]) zoutf<<_geno_dose[_keep[i]][_include[j]]<<' ';
else zoutf<<2.0-_geno_dose[_keep[i]][_include[j]]<<' ';
}
else{
if(miss_with_mu) zoutf<<_mu[_include[j]]<<' ';
else zoutf<<"NA ";
}
}
}
else{
for(j=0; j<_include.size(); j++){
if(!_snp_1[_include[j]][_keep[i]] || _snp_2[_include[j]][_keep[i]]){
if(_allele1[_include[j]]==_ref_A[_include[j]]) zoutf<<_snp_1[_include[j]][_keep[i]]+_snp_2[_include[j]][_keep[i]]<<' ';
else zoutf<<2.0-(_snp_1[_include[j]][_keep[i]]+_snp_2[_include[j]][_keep[i]])<<' ';
}
else{
if(miss_with_mu) zoutf<<_mu[_include[j]]<<' ';
else zoutf<<"NA ";
}
}
}
zoutf<<endl;
}
zoutf.close();
cout<<"The recoded genotype matrix has been saved in the file ["+X_zFile+"] (in compressed text format) ."<<endl;
}
void gcta::save_freq(bool ssq_flag)
{
if(_mu.empty()) calcu_mu(ssq_flag);
string save_freq=_out+".freq";
ofstream ofreq(save_freq.c_str());
if(!ofreq) throw("Error: can not open the file ["+save_freq+"] to write.");
int i=0;
cout<<"Writing allele frequencies of "<<_include.size()<<" SNPs to ["+save_freq+"]."<<endl;
for(i=0; i<_include.size(); i++){
ofreq<<_snp_name[_include[i]]<<"\t"<<_ref_A[_include[i]]<<"\t"<<setprecision(15)<<_mu[_include[i]]*0.5;
// if(ssq_flag) ofreq<<"\t"<<_ssq[_include[i]]<<"\t"<<_w[_include[i]];
ofreq<<endl;
}
ofreq.close();
cout<<"Allele frequencies of "<<_include.size()<<" SNPs have been saved in the file ["+save_freq+"]."<<endl;
}
void gcta::std_XMat(vector< vector<float> > &X, vector<double> &sd_SNP, bool grm_xchr_flag, bool divid_by_std)
{
if(_mu.empty()) calcu_mu();
int i=0, j=0;
sd_SNP.clear();
sd_SNP.resize(_include.size()); // SD of each SNP, sqrt(2pq)
if(_dosage_flag){
for(j=0; j<_include.size(); j++){
for(i=0; i<_keep.size(); i++) sd_SNP[j]+=(X[i][j]-_mu[_include[j]])*(X[i][j]-_mu[_include[j]]);
sd_SNP[j]/=(_keep.size()-1.0);
}
}
else{
for(j=0; j<_include.size(); j++) sd_SNP[j]=_mu[_include[j]]*(1.0-0.5*_mu[_include[j]]);
}
for(j=0; j<_include.size(); j++){
if(fabs(sd_SNP[j])<1.0e-50) sd_SNP[j]=0.0;
else sd_SNP[j]=sqrt(1.0/sd_SNP[j]);
}
for(i=0; i<_keep.size(); i++){
for(j=0; j<_include.size(); j++){
if(X[i][j]<1e5){
X[i][j]-=_mu[_include[j]];
if(divid_by_std) X[i][j]*=sd_SNP[j];
}
}
}
if(!grm_xchr_flag) return;
// for the X-chromosome
check_sex();
double f_buf=sqrt(0.5);
for(i=0; i<_keep.size(); i++){
if(_sex[_keep[i]]==1){
for(j=0; j<_include.size(); j++){
if(X[i][j]<1e5) X[i][j]*=f_buf;
}
}
}
}
void gcta::filter_snp_max_maf(double max_maf)
{
if(_mu.empty()) calcu_mu();
cout<<"Pruning SNPs with MAF < "<<max_maf<<" ..."<<endl;
map<string, int> id_map_buf(_snp_name_map);
map<string, int>::iterator iter, end=id_map_buf.end();
int prev_size=_include.size();
double fbuf=0.0;
_include.clear();
_snp_name_map.clear();
for(iter=id_map_buf.begin(); iter!=end; iter++){
fbuf=_mu[iter->second]*0.5;
if(fbuf>max_maf && 1.0-fbuf>max_maf) continue;
_snp_name_map.insert(*iter);
_include.push_back(iter->second);
}
if(_include.size()==0) throw("Error: No SNP is retained for analysis.");
else{
stable_sort(_include.begin(), _include.end());
cout<<"After pruning SNPs with MAF < "<<max_maf<<", there are "<<_include.size()<<" SNPs ("<<prev_size-_include.size()<<" SNPs with MAF > "<<max_maf<<")."<<endl;
}
}
void gcta::make_XMat(vector< vector<float> > &X, bool miss_with_mu)
{
if(_mu.empty() && miss_with_mu) calcu_mu();
cout<<"Recoding genotypes (individual major mode) ..."<<endl;
int i=0, j=0;
X.clear();
X.resize(_keep.size());
for(i=0; i<_keep.size(); i++){
X[i].resize(_include.size());
bool need2fill=false;
if(_dosage_flag){
for(j=0; j<_include.size(); j++){
if(_geno_dose[_keep[i]][_include[j]]<1e5){
if(_allele1[_include[j]]==_ref_A[_include[j]]) X[i][j]=_geno_dose[_keep[i]][_include[j]];
else X[i][j]=2.0-_geno_dose[_keep[i]][_include[j]];
}
else{ X[i][j]=1e6; need2fill=true; }
}
_geno_dose[i].clear();
}
else{
for(j=0; j<_include.size(); j++){
if(!_snp_1[_include[j]][_keep[i]] || _snp_2[_include[j]][_keep[i]]){
if(_allele1[_include[j]]==_ref_A[_include[j]]) X[i][j]=_snp_1[_include[j]][_keep[i]]+_snp_2[_include[j]][_keep[i]];
else X[i][j]=2.0-(_snp_1[_include[j]][_keep[i]]+_snp_2[_include[j]][_keep[i]]);
}
else{ X[i][j]=1e6; need2fill=true; }
}
}
// Fill the missing genotype with the mean of x (2p)
for(j=0; j<_include.size() && miss_with_mu && need2fill; j++){
if(X[i][j]>1e5) X[i][j]=_mu[_include[j]];
}
}
}
