def univariate(snps, X, Y, col=pheno_name):
p = X.shape[1]
from scipy.stats import pearsonr
pvals = []
cors = []
for i in range(X.shape[1]):
cor, pval = pearsonr(X[:, i], Y)
cors.append(cor)
pvals.append(pval)
pvals = np.asarray(pvals)
cors = np.asarray(cors)
indices = np.where(pvals <= 0.05)
print "\n"
print "..... Univariate results"
print ' numbers of significant p values *un*corrected', len(
indices[0]), 'over ', p
import p_value_correction as p_c
p_corrected = p_c.fdr(pvals)
w = np.where(p_corrected <= 0.05)[0]
print ' numbers of significant corrected p values corrected', len(
w), 'over ', p
print ' ', snps.measure_ids[w], " pvalcor = ", p_corrected[
w], " correlation = ", cors[w]
"""lm = LinearRegression()
p = X.shape[1]
#the p_values computation for the univariate approech
from scipy.stats import pearsonr
p_vect = np.array([])
cor_vect = np.array([])
for i in range(X.shape[1]):
r_row, p_value = pearsonr(X[:, i], Y)
p_vect = np.hstack((p_vect, p_value))
cor_vect = np.hstack((cor_vect, r_row))
indices = np.where(p_vect <= 0.05)
print 'numbers of significant p values', len(indices[0]), 'over ', p
#correction of the p_values using the fdr approech
import p_value_correction as p_c
p_corrected = p_c.fdr(p_vect)
indices_c = np.where(p_corrected <= 0.05)
print 's0 : numbers of significant corrected p values', len(
indices_c[0]), 'over ', p
import matplotlib.pyplot as plt
plt.figure(2)
plt.subplot(211)
plt.hist(p_corrected, 40)
plt.title('corrected p values ')
plt.subplot(212)
plt.hist(p_vect, 40)
plt.title('uncorrected p values ')
plt.show()
import matplotlib.pyplot as plt
def univariate(mask, snps, studyPgS, col='height'):
# the SNP are X and reordered lines
X = snps.data[mask, :]
p = X.shape[1]
permuter = snps.subject_ids[mask].tolist()
y = studyPgS.loc[permuter][col]
covariate = numpy.matrix(
pandas.get_dummies(studyPgS.loc[permuter]['ScanningCentre'],
prefix='Centre')[range(7)])
print "COVARIATE"
print covariate
covariate = numpy.hstack(
(covariate, numpy.asarray(studyPgS.loc[permuter][['Sex', 'Age']])))
print "COVARIATE"
print covariate
from sklearn.linear_model import LinearRegression
Y = y - LinearRegression().fit(covariate, y).predict(covariate)
from scipy.stats import pearsonr
pvals = []
cors = []
for i in range(X.shape[1]):
cor, pval = pearsonr(X[:, i], Y)
cors.append(cor)
pvals.append(pval)
pvals = numpy.asarray(pvals)
cors = numpy.asarray(cors)
indices = numpy.where(pvals <= 0.05)
print "\n"
print "..... Univariate results"
print ' numbers of significant p values *un*corrected', len(
indices[0]), 'over ', p
import p_value_correction as p_c
p_corrected = p_c.fdr(pvals)
w = numpy.where(p_corrected <= 0.05)[0]
print ' numbers of significant corrected p values corrected', len(
w), 'over ', p
print ' ', snps.measure_ids[w], " pvalcor = ", p_corrected[
w], " correlation = ", cors[w]
if col == 'height':
snps_mask = [
snps.measure_ids.tolist().index(i) for i in snps.measure_ids[w]
]
subX = snps.data[mask, :][:, snps_mask]
lm = LinearRegression()
lm.fit(subX, Y)
print "\n..... Score explained by the %d significant SNPS is ~ 1.5 percent of the height var" % len(
w)
print " based on: ", subX.shape[0], ' subjects'
print " covariate out is sex , age, scanning center"
print 'lm.score(X, Y)', lm.score(subX, Y)
else:
if len(w) > 0:
print "\n..... Score explained by the %d significant SNPS of the var" % (
len(w), col)
print " based on: ", subX.shape[0], ' subjects'
print " covariate out is sex , age, scanning center"
print 'lm.score(X, Y)', lm.score(subX, Y)
else:
print "\n..... Nothing in %s variability explained by this approach " % (
col)
return X, Y
plt.subplot(221)
plt.hist(p_vect_sex_0, 20)
plt.title('uncorrected p values for sex 0')
plt.subplot(222)
plt.hist(p_vect_sex_1, 20)
plt.title('uncorrected p values for sex 1')
plt.subplot(223)
plt.plot(cor_vect_sex_0)
plt.title('correlation p coef sex 0')
plt.subplot(224)
plt.plot(cor_vect_sex_1)
plt.title('correlation coef for sex 1')
plt.show()
import p_value_correction as p_c
p_corrected_sex_0 = p_c.fdr(p_vect_sex_0)
indices_c_sex_0 = np.where(p_corrected_sex_0 <= 0.05)
p_corrected_sex_1 = p_c.fdr(p_vect_sex_1)
indices_c_sex_1 = np.where(p_corrected_sex_1 <= 0.05)
print 's0 : numbers of significant corrected p values', len(
indices_c_sex_0[0]), 'over ', X_.shape[1]
print 's1 :numbers of significant corrected p values', len(
indices_c_sex_1[0]), 'over ', X_.shape[1]
plt.figure(2)
plt.subplot(211)
plt.hist(p_corrected_sex_0, 20)
plt.title('corrected p values for sex 0')
indices_res = np.where(p_vect_res <= 0.05)
print 'numbers of significant p values', len(indices_res[0]), 'over ', p
#
plt.figure(4)
plt.subplot(211)
plt.hist(p_vect_res, 20)
plt.title('uncorrected p values ')
plt.subplot(212)
plt.plot(cor_vect_res)
plt.title('correlation p coef')
plt.show()
import p_value_correction as p_c
p_corrected_res = p_c.fdr(p_vect_res)
indices_c_res = np.where(p_corrected_res <= 0.05)
print 'numbers of significant corrected p values', len(
indices_c_res[0]), 'over ', p
plt.figure(5)
plt.hist(p_corrected_res, 20)
plt.show()
from sklearn.linear_model import LinearRegression
from sklearn.utils import check_random_state
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_regression
import itertools
import operator
cor_vect_eig_SNP = np.array([])
p_eig_SNP = X_new.shape[1]
for i in range(p_eig_SNP):
r_row_eig_SNP, p_value_eig_SNP = pearsonr(X_new[:, i], y)
p_vect_eig_SNP = np.hstack((p_vect_eig_SNP, p_value_eig_SNP))
cor_vect_eig_SNP = np.hstack((cor_vect_eig_SNP, r_row_eig_SNP))
indices_eig_SNP = np.where(p_vect_eig_SNP <= 0.05)
print 'numbers of significant p values for _eig_SNP', len(
indices_eig_SNP[0]), 'over ', p_eig_SNP
plt.figure(1)
plt.subplot(211)
plt.hist(p_vect_eig_SNP, 20)
plt.title('uncorrected p values _eig_SNP ')
plt.subplot(212)
plt.plot(cor_vect_eig_SNP)
plt.title('correlation p coef _eig_SNP')
plt.show()
import p_value_correction as p_c
p_corrected_eig_SNP = p_c.fdr(p_vect_eig_SNP)
indices_c_eig_SNP = np.where(p_corrected_eig_SNP <= 0.05)
print 's0 : numbers of significant corrected p values', len(
indices_c_eig_SNP[0]), 'over ', X_new.shape[1]
plt.figure(2)
plt.hist(p_corrected_eig_SNP)
plt.title('corrected p values for _eig_SNP')
plt.show()