def TESTBEFOREUSING_pca_covariates(bedfile, npc, outfile=None):
'''
Read in bed file and compute PC covariates (a la Eigenstrat)
bedfile should include the .bed extension
returns pccov, U, S
'''
import fastlmm.pyplink.plink as pl
import os.path
import time
import scipy.linalg as la
import numpy as np
root, ext = os.path.splitext(bedfile)
print "reading in bed file..."
t0 = time.time()
SNPs = Bed(root).read().standardize()
t1 = time.time()
print("Elapsed time %.2f seconds" % (t1 - t0))
[N, M] = snps.shape
print "found nind=" + str(N) + ", nsnp=" + str(M)
if M < N:
print "only doing full rank, should use low rank here"
#not needed, and in practice, makes no difference
#mean center the individuals
#meanval=sp.mean(snps,axis=0)
#snps=snps-meanval
print "computing kernel..."
t0 = time.time()
K = sp.dot(snps, snps.T)
t1 = time.time()
print("Elapsed time %.2f seconds" % (t1 - t0))
snps = None
t0 = time.time()
print "computing svd..."
[U, S, V] = la.svd(K, full_matrices=False)
t1 = time.time()
print("Elapsed time %.2f seconds" % (t1 - t0))
S = sp.sqrt(S)
#UShalf=sp.dot(U,sp.diag(S)) #expensive
UShalf = sp.multiply(U, np.tile(
S,
(N,
1))) #faster, but not as fast as as_strided which I can't get to work
pccov = UShalf[:, 0:npc]
print "done."
if outfile is not None:
import fastlmm.util.util as ut
ut.write_plink_covariates(SNPs['iid'], pccov, outfile)
return pccov, U, S
def TESTBEFOREUSING_pca_covariates(bedfile,npc,outfile=None):
'''
Read in bed file and compute PC covariates (a la Eigenstrat)
bedfile should include the .bed extension
returns pccov, U, S
'''
import fastlmm.pyplink.plink as pl
import os.path
import time
import scipy.linalg as la
import numpy as np
root, ext = os.path.splitext(bedfile)
print "reading in bed file..."
t0=time.time()
SNPs = Bed(root).read().standardize()
t1=time.time()
print ("Elapsed time %.2f seconds" % (t1-t0))
[N,M]=snps.shape
print "found nind=" + str(N) + ", nsnp=" + str(M)
if M<N:
print "only doing full rank, should use low rank here"
#not needed, and in practice, makes no difference
#mean center the individuals
#meanval=sp.mean(snps,axis=0)
#snps=snps-meanval
print "computing kernel..."
t0=time.time()
K=sp.dot(snps,snps.T)
t1=time.time()
print ("Elapsed time %.2f seconds" % (t1-t0))
snps=None
t0=time.time()
print "computing svd..."
[U,S,V] = la.svd(K,full_matrices = False)
t1=time.time()
print ("Elapsed time %.2f seconds" % (t1-t0))
S=sp.sqrt(S)
#UShalf=sp.dot(U,sp.diag(S)) #expensive
UShalf=sp.multiply(U,np.tile(S,(N,1))) #faster, but not as fast as as_strided which I can't get to work
pccov=UShalf[:,0:npc]
print "done."
if outfile is not None:
import fastlmm.util.util as ut
ut.write_plink_covariates(SNPs['iid'],pccov,outfile)
return pccov,U,S
def linreg_entire_dataset(self, output_prefix):
self.run_once()
# We do a little extra work here. We write the pcs out here because they will be known here and we know this method will only be run
# once even on a big cluster jobs
if output_prefix is not None and self.num_pcs > 0:
filename = self.pcs_filename(output_prefix)
import fastlmm.util.util as ut
ut.write_plink_covariates(self.ind_iid, self.pcs, filename)
logging.info("performing final scan through entire data set")
tt0 = time.time()
_F, _pval = self.G.lin_reg(self.y, self.X)
feat_idx = np.argsort(_pval)
_pval_feat_idx = _pval[feat_idx]
sid_feat_idx = self.sid[feat_idx]
lingreg_results = (_pval[feat_idx], self.sid[feat_idx])
logging.info("fin_scan time %.2f s" % (float(time.time() - tt0)))
return lingreg_results
def linreg_entire_dataset(self, output_prefix):
self.run_once()
# We do a little extra work here. We write the pcs out here because they will be known here and we know this method will only be run
# once even on a big cluster jobs
if output_prefix is not None and self.num_pcs > 0:
filename = self.pcs_filename(output_prefix)
import fastlmm.util.util as ut
ut.write_plink_covariates(self.ind_iid,self.pcs,filename)
logging.info("performing final scan through entire data set")
tt0 = time.time()
_F,_pval = self.G.lin_reg(self.y, self.X)
feat_idx = np.argsort(_pval)
_pval_feat_idx = _pval[feat_idx]
sid_feat_idx = self.sid[feat_idx]
lingreg_results = (_pval[feat_idx],self.sid[feat_idx])
logging.info("fin_scan time %.2f s" % (float(time.time() - tt0)))
return lingreg_results
