def testLassoBackgroundModel(estimator, Xfg, Xbg, y, use_Kgeno=True, **kwargs):
"""
run a linear mixed model on the foreground SNPs (Xfg), while estimating SNPs for the background covariance matrix via the Lasso usting the background SNPs(Xbg).
estimator : (Lmm-)Lasso model
Xbg : SNP data for the background model
Xfg : SNP data for the foreground model
y : phenotype vector
use_Kgene : if True (default), background covariance matrix is estimated and used.
"""
if use_Kgeno:
Kbg = compute_linear_kernel(Xbg)
estimator.fit(Xfg, y, Kbg)
iactive = estimator.coef_ != 0
if iactive.any():
Kfg = compute_linear_kernel(Xbg, iactive)
vd = VarianceDecomposition(y)
vd.addRandomEffect(is_noise=True)
vd.addRandomEffect(Kbg)
vd.addRandomEffect(Kfg)
vd.optimize()
K = vd.gp.getCovar().K()
else:
K = Kbg
else:
estimator.fit(Xbg, y)
idx = estimator.coef_ != 0
K = compute_linear_kernel(Xbg, idx)
lmm = qtl.test_lmm(Xfg, y, K=K, **kwargs)
pv = lmm.getPv()
return pv
def fit(self, X, y, K, standardize=False, verbose=False, **lasso_args):
"""
fitting the model
X: SNP data
y: phenotype data
K: backgroundcovariance matrix
standardize: if True, genotypes and phenotypes are standardized
"""
if y.ndim == 2:
assert y.shape[
1] == 1, 'Only one phenotype can be processed at at time.'
y = y.flatten()
time_start = time.time()
[n_s, n_f] = X.shape
assert X.shape[0] == y.shape[0], 'dimensions do not match'
assert K.shape[0] == K.shape[1], 'dimensions do not match'
assert K.shape[0] == X.shape[0], 'dimensions do not match'
""" standardizing genotypes and phenotypes """
if standardize:
X -= X.mean(axis=0)
X /= X.std(axis=0)
y -= y.mean(axis=0)
y /= y.std(axis=0)
""" training null model """
vd = VarianceDecomposition(y)
vd.addRandomEffect(is_noise=True)
vd.addRandomEffect(K)
vd.optimize()
varComps = vd.getVarianceComps()
delta0 = varComps[0, 0] / varComps.sum()
self.varComps = varComps
S, U = LA.eigh(K)
""" rotating data """
Sdi = 1. / (S + delta0)
Sdi_sqrt = SP.sqrt(Sdi)
SUX = SP.dot(U.T, X)
SUX = SUX * SP.tile(Sdi_sqrt, (n_f, 1)).T
SUy = SP.dot(U.T, y)
SUy = Sdi_sqrt * SUy
""" fitting lasso """
super(LmmLasso, self).fit(SUX, SUy, **lasso_args)
yhat = super(LmmLasso, self).predict(X)
self.w_ridge = LA.solve(K + delta0 * SP.eye(n_s), y - yhat)
time_end = time.time()
time_diff = time_end - time_start
if verbose: print '... finished in %.2fs' % (time_diff)
return self
def testLassoBackgroundModel(estimator,Xfg,Xbg,y,use_Kgeno=True,**kwargs):
"""
run a linear mixed model on the foreground SNPs (Xfg), while estimating SNPs for the background covariance matrix via the Lasso usting the background SNPs(Xbg).
estimator : (Lmm-)Lasso model
Xbg : SNP data for the background model
Xfg : SNP data for the foreground model
y : phenotype vector
use_Kgene : if True (default), background covariance matrix is estimated and used.
"""
if use_Kgeno:
Kbg = compute_linear_kernel(Xbg)
estimator.fit(Xfg,y,Kbg)
iactive = estimator.coef_!=0
if iactive.any():
Kfg = compute_linear_kernel(Xbg,iactive)
vd = VarianceDecomposition(y)
vd.addRandomEffect(is_noise=True)
vd.addRandomEffect(Kbg)
vd.addRandomEffect(Kfg)
vd.optimize()
K = vd.gp.getCovar().K()
else:
K = Kbg
else:
estimator.fit(Xbg,y)
idx = estimator.coef_!=0
K = compute_linear_kernel(Xbg,idx)
lmm = qtl.test_lmm(Xfg,y,K=K,**kwargs)
pv = lmm.getPv()
return pv
def fit(self, X, y, K, standardize=False, verbose=False,**lasso_args):
"""
fitting the model
X: SNP data
y: phenotype data
K: backgroundcovariance matrix
standardize: if True, genotypes and phenotypes are standardized
"""
if y.ndim == 2:
assert y.shape[1]==1, 'Only one phenotype can be processed at at time.'
y = y.flatten()
time_start = time.time()
[n_s, n_f] = X.shape
assert X.shape[0] == y.shape[0], 'dimensions do not match'
assert K.shape[0] == K.shape[1], 'dimensions do not match'
assert K.shape[0] == X.shape[0], 'dimensions do not match'
""" standardizing genotypes and phenotypes """
if standardize:
X -= X.mean(axis=0)
X /= X.std(axis=0)
y -= y.mean(axis=0)
y /= y.std(axis=0)
""" training null model """
vd = VarianceDecomposition(y)
vd.addRandomEffect(is_noise=True)
vd.addRandomEffect(K)
vd.optimize()
varComps = vd.getVarianceComps()
delta0 = varComps[0,0]/varComps.sum()
self.varComps = varComps
S,U = LA.eigh(K)
""" rotating data """
Sdi = 1. / (S + delta0)
Sdi_sqrt = SP.sqrt(Sdi)
SUX = SP.dot(U.T, X)
SUX = SUX * SP.tile(Sdi_sqrt, (n_f, 1)).T
SUy = SP.dot(U.T, y)
SUy = Sdi_sqrt * SUy
""" fitting lasso """
super(LmmLasso, self).fit(SUX, SUy, **lasso_args)
yhat = super(LmmLasso, self).predict(X)
self.w_ridge = LA.solve(K + delta0 * SP.eye(n_s), y - yhat)
time_end = time.time()
time_diff = time_end - time_start
if verbose: print '... finished in %.2fs'%(time_diff)
return self