def gwas(bedSim, bedTest, pheno, h2, outFile, eigenFile, covar): bedSim, pheno = leapUtils._fixupBedAndPheno(bedSim, pheno) bedTest, pheno = leapUtils._fixupBedAndPheno(bedTest, pheno) #Run GWAS logdelta = np.log(1.0/h2 - 1) G0 = (bedSim if eigenFile is None else None) print 'Performing LEAP GWAS...' results_df = fastlmm.association.single_snp(bedTest, pheno, G0=G0, covar=covar, output_file_name=outFile, log_delta=logdelta, cache_file=eigenFile) return results_df
def gwas(bedSim, bedTest, pheno, h2, outFile, eigenFile, covar): bedSim, pheno = leapUtils._fixupBedAndPheno(bedSim, pheno) bedTest, pheno = leapUtils._fixupBedAndPheno(bedTest, pheno) #Run GWAS logdelta = np.log(1.0/h2 - 1) G0 = (bedSim if eigenFile is None else None) print 'Performing LEAP GWAS...' results_df = fastlmm.association.single_snp(bedTest, pheno, G0=G0, covar=covar, output_file_name=outFile, log_delta=logdelta, cache_file=eigenFile) return results_df
def probit(bed,
pheno,
h2,
prev,
eigen,
outFile,
keepArr,
covar,
thresholds,
nofail,
numSkipTopPCs,
mineig,
hess,
recenter,
maxFixedIters,
epsilon,
treatFixedAsRandom=False):
bed, pheno = leapUtils._fixupBedAndPheno(bed, pheno)
#Extract phenotype
if isinstance(pheno, dict): phe = pheno['vals']
else: phe = pheno
if (len(phe.shape) == 2):
if (phe.shape[1] == 1): phe = phe[:, 0]
else: raise Exception('More than one phenotype found')
if (keepArr is None): keepArr = np.ones(phe.shape[0], dtype=np.bool)
S = eigen['arr_1'] * bed.sid.shape[0]
U = eigen['arr_0']
S = np.sqrt(S)
goodS = (S > mineig)
if (numSkipTopPCs > 0): goodS[-numSkipTopPCs:] = False
if (np.sum(~goodS) > 0):
print 'Removing', np.sum(~goodS), 'PCs with low variance'
G = U[:, goodS] * S[goodS]
#Set binary vector
pheUnique = np.unique(phe)
if (pheUnique.shape[0] != 2):
raise Exception('phenotype file has more than two values')
pheMean = phe.mean()
cases = (phe > pheMean)
phe[~cases] = 0
phe[cases] = 1
#run probit regression
t = stats.norm(0, 1).isf(prev)
if (thresholds is not None): t = thresholds
#Recenter G to only consider the unrelated individuals
if recenter: G -= np.mean(G[keepArr, :], axis=0)
else: G -= np.mean(G, axis=0)
numFixedFeatures = 0
if (covar is not None):
covar -= covar.mean()
covar /= covar.std()
covar *= np.mean(np.std(G, axis=0))
G = np.concatenate((covar, G), axis=1)
if (not treatFixedAsRandom): numFixedFeatures += covar.shape[1]
#Run Probit regression
probitThresh = (t if thresholds is None else t[keepArr])
beta = probitRegression(G[keepArr, :], phe[keepArr], probitThresh,
bed.sid.shape[0], numFixedFeatures, h2, hess,
maxFixedIters, epsilon, nofail)
#Predict liabilities for all individuals
meanLiab = G.dot(beta)
liab = meanLiab.copy()
indsToFlip = ((liab <= t) & (phe > 0.5)) | ((liab > t) & (phe < 0.5))
liab[indsToFlip] = stats.norm(0, 1).isf(prev)
if (outFile is not None):
#save liabilities
f = open(outFile + '.liabs', 'w')
for ind_i, [fid, iid] in enumerate(bed.iid):
f.write(' '.join([fid, iid, '%0.3f' % liab[ind_i]]) + '\n')
f.close()
#save liabilities after regressing out the fixed effects
if (numFixedFeatures > 0):
liab_nofixed = liab - G[:, :numFixedFeatures].dot(
beta[:numFixedFeatures])
f = open(outFile + '.liab_nofixed', 'w')
for ind_i, [fid, iid] in enumerate(bed.iid):
f.write(' '.join([fid, iid,
'%0.3f' % liab_nofixed[ind_i]]) + '\n')
f.close()
liab_nofixed2 = meanLiab - G[:, :numFixedFeatures].dot(
beta[:numFixedFeatures])
indsToFlip = ((liab_nofixed2 <= t) &
(phe > 0.5)) | ((liab_nofixed2 > t) & (phe < 0.5))
liab_nofixed2[indsToFlip] = stats.norm(0, 1).isf(prev)
f = open(outFile + '.liab_nofixed2', 'w')
for ind_i, [fid, iid] in enumerate(bed.iid):
f.write(' '.join([fid, iid,
'%0.3f' % liab_nofixed2[ind_i]]) + '\n')
f.close()
#Return phenotype struct with liabilities
liabsStruct = {'header': [None], 'vals': liab, 'iid': bed.iid}
return liabsStruct
def probit(bed, pheno, h2, prev, eigen, outFile, keepArr, covar, thresholds, nofail,
numSkipTopPCs, mineig, hess, recenter, maxFixedIters, epsilon, treatFixedAsRandom=False):
bed, pheno = leapUtils._fixupBedAndPheno(bed, pheno)
#Extract phenotype
if isinstance(pheno, dict): phe = pheno['vals']
else: phe = pheno
if (len(phe.shape)==2):
if (phe.shape[1]==1): phe=phe[:,0]
else: raise Exception('More than one phenotype found')
if (keepArr is None): keepArr = np.ones(phe.shape[0], dtype=np.bool)
S = eigen['arr_1'] * bed.sid.shape[0]
U = eigen['arr_0']
S = np.sqrt(S)
goodS = (S>mineig)
if (numSkipTopPCs > 0): goodS[-numSkipTopPCs:] = False
if (np.sum(~goodS) > 0): print 'Removing', np.sum(~goodS), 'PCs with low variance'
G = U[:, goodS]*S[goodS]
#Set binary vector
pheUnique = np.unique(phe)
if (pheUnique.shape[0] != 2): raise Exception('phenotype file has more than two values')
pheMean = phe.mean()
cases = (phe>pheMean)
phe[~cases] = 0
phe[cases] = 1
#run probit regression
t = stats.norm(0,1).isf(prev)
if (thresholds is not None): t = thresholds
#Recenter G to only consider the unrelated individuals
if recenter: G -= np.mean(G[keepArr, :], axis=0)
else: G -= np.mean(G, axis=0)
numFixedFeatures = 0
if (covar is not None):
covar -= covar.mean()
covar /= covar.std()
covar *= np.mean(np.std(G, axis=0))
G = np.concatenate((covar, G), axis=1)
if (not treatFixedAsRandom): numFixedFeatures += covar.shape[1]
#Run Probit regression
probitThresh = (t if thresholds is None else t[keepArr])
beta = probitRegression(G[keepArr, :], phe[keepArr], probitThresh, bed.sid.shape[0], numFixedFeatures, h2, hess, maxFixedIters, epsilon, nofail)
#Predict liabilities for all individuals
meanLiab = G.dot(beta)
liab = meanLiab.copy()
indsToFlip = ((liab <= t) & (phe>0.5)) | ((liab > t) & (phe<0.5))
liab[indsToFlip] = stats.norm(0,1).isf(prev)
if (outFile is not None):
#save liabilities
f = open(outFile+'.liabs', 'w')
for ind_i,[fid,iid] in enumerate(bed.iid): f.write(' '.join([fid, iid, '%0.3f'%liab[ind_i]]) + '\n')
f.close()
#save liabilities after regressing out the fixed effects
if (numFixedFeatures > 0):
liab_nofixed = liab - G[:, :numFixedFeatures].dot(beta[:numFixedFeatures])
f = open(outFile+'.liab_nofixed', 'w')
for ind_i,[fid,iid] in enumerate(bed.iid): f.write(' '.join([fid, iid, '%0.3f'%liab_nofixed[ind_i]]) + '\n')
f.close()
liab_nofixed2 = meanLiab - G[:, :numFixedFeatures].dot(beta[:numFixedFeatures])
indsToFlip = ((liab_nofixed2 <= t) & (phe>0.5)) | ((liab_nofixed2 > t) & (phe<0.5))
liab_nofixed2[indsToFlip] = stats.norm(0,1).isf(prev)
f = open(outFile+'.liab_nofixed2', 'w')
for ind_i,[fid,iid] in enumerate(bed.iid): f.write(' '.join([fid, iid, '%0.3f'%liab_nofixed2[ind_i]]) + '\n')
f.close()
#Return phenotype struct with liabilities
liabsStruct = {
'header':[None],
'vals':liab,
'iid':bed.iid
}
return liabsStruct
