def __init__(self, z1, z2, x, w, N1, N2, M, intercept_hsq1=None, intercept_hsq2=None,
intercept_gencov=None, n_blocks=200, slow=False, twostep=None):
self.intercept_gencov = intercept_gencov
self._negative_hsq = None
n_snp, n_annot = x.shape
hsq1 = Hsq(np.square(z1), x, w, N1, M, n_blocks=n_blocks, intercept=intercept_hsq1,
slow=slow, twostep=twostep)
hsq2 = Hsq(np.square(z2), x, w, N2, M, n_blocks=n_blocks, intercept=intercept_hsq2,
slow=slow, twostep=twostep)
gencov = Gencov(z1, z2, x, w, N1, N2, M, hsq1.tot, hsq2.tot, hsq1.intercept,
hsq2.intercept, n_blocks, intercept_gencov=intercept_gencov, slow=slow,
twostep=twostep)
gencov.N1 = None # save memory
gencov.N2 = None
self.hsq1, self.hsq2, self.gencov = hsq1, hsq2, gencov
if (hsq1.tot <= 0 or hsq2.tot <= 0):
self._negative_hsq = True
self.rg_ratio = self.rg = self.rg_se = 'NA'
self.p = self.z = 'NA'
else:
rg_ratio = np.array(
gencov.tot / np.sqrt(hsq1.tot * hsq2.tot)).reshape((1, 1))
denom_delete_values = np.sqrt(
np.multiply(hsq1.tot_delete_values, hsq2.tot_delete_values))
rg = jk.RatioJackknife(
rg_ratio, gencov.tot_delete_values, denom_delete_values)
self.rg_jknife = float(rg.jknife_est)
self.rg_se = float(rg.jknife_se)
self.rg_ratio = float(rg_ratio)
self.p, self.z = p_z_norm(self.rg_ratio, self.rg_se)
def _enrichment_g(self,
g_overlap_matrix,
coef,
ave_coef,
jknife,
n_annot,
n_g_tot_annot,
indices=None):
n_blocks = jknife.delete_values.shape[0]
delete_vals = np.dot(g_overlap_matrix,
jknife.delete_values[:, n_annot:self.n_annot].T).T
if indices:
numer_delete_vals = np.mean(delete_vals[:, indices],
axis=1).reshape((n_blocks, 1))
else:
numer_delete_vals = delete_vals
denom_delete_vals = np.mean(delete_vals[:, :n_g_tot_annot],
axis=1).reshape((n_blocks, 1))
if not indices:
denom_delete_vals = np.dot(denom_delete_vals,
np.ones((1, self.n_annot - n_annot)))
prop = jk.RatioJackknife(coef / ave_coef, numer_delete_vals,
denom_delete_vals)
return prop.est, prop.jknife_cov, prop.jknife_se
def _prop(self, jknife, M, Nbar, cat, tot):
'''Convert total h2 and per-category h2 to per-category proportion h2 or gencov.'''
n_annot = self.n_annot
n_blocks = jknife.delete_values.shape[0]
numer_delete_vals = np.multiply(
M, jknife.delete_values[:, 0:n_annot]) / Nbar # (n_blocks, n_annot)
denom_delete_vals = np.sum(
numer_delete_vals, axis=1).reshape((n_blocks, 1))
denom_delete_vals = np.dot(denom_delete_vals, np.ones((1, n_annot)))
prop = jk.RatioJackknife(
cat / tot, numer_delete_vals, denom_delete_vals)
return prop.est, prop.jknife_cov, prop.jknife_se
def _prop(self, jknife, M, Nbar, cat, tot, use_bootstrap=False):
'''Convert total h2 and per-category h2 to per-category proportion h2 or gencov.'''
n_annot = self.n_annot
n_blocks = jknife.delete_values.shape[0]
numer_delete_vals = np.multiply(
M, jknife.delete_values[:, 0:n_annot]) / Nbar # (n_blocks, n_annot)
denom_delete_vals = np.sum(
numer_delete_vals, axis=1).reshape((n_blocks, 1))
denom_delete_vals = np.dot(denom_delete_vals, np.ones((1, n_annot)))
if use_bootstrap:
bootstrap_ratios = numer_delete_vals / denom_delete_vals
###prop_est = np.mean(bootstrap_ratios, axis=0)
prop_est = cat / tot
prop_cov = np.cov(bootstrap_ratios.T, ddof=1)
prop_se = np.sqrt(np.diag(prop_cov))
return prop_est, prop_cov, prop_se
else:
prop = jk.RatioJackknife(
cat / tot, numer_delete_vals, denom_delete_vals)
return prop.est, prop.jknife_cov, prop.jknife_se