# In[18]:
stats = pd.DataFrame(index=phenos, columns=['test_stat', 'pval'])
for i, pheno in enumerate(phenos):
x = df.loc[df.loc[:, 'sex'] == 1, pheno]
# x = df.loc[np.logical_and(df[train_test_str] == 1,df['sex'] == 1),pheno]
y = df.loc[df.loc[:, 'sex'] == 2, pheno]
# y = df.loc[np.logical_and(df[train_test_str] == 1,df['sex'] == 2),pheno]
test_output = sp.stats.ttest_ind(x, y)
stats.loc[pheno, 'test_stat'] = test_output[0]
stats.loc[pheno, 'pval'] = test_output[1]
stats.loc[:, 'pval_corr'] = get_fdr_p(stats.loc[:, 'pval'])
stats.loc[:, 'sig'] = stats.loc[:, 'pval_corr'] < 0.05
np.round(stats.astype(float), 2)
# In[19]:
f, ax = plt.subplots(1, len(phenos))
f.set_figwidth(len(phenos) * 1.4)
f.set_figheight(1.25)
# sex: 1=male, 2=female
for i, pheno in enumerate(phenos):
x = df.loc[df.loc[:, 'sex'] == 1, pheno]
# x = df.loc[np.logical_and(df[train_test_str] == 1,df['sex'] == 1),pheno]
sns.kdeplot(x, ax=ax[i], label='male', color='b')
assign_p=assign_p,
nulldir=nulldir)
elif assign_p == 'parametric':
df_pheno_z = run_pheno_correlations(df.loc[:, phenos],
df_z,
method=method,
assign_p=assign_p)
# In[22]:
# correct multiple comparisons. We do this across brain regions and phenotypes (e.g., 400*6 = 2400 tests)
df_p_corr = pd.DataFrame(index=df_pheno_z.index,
columns=['p-corr']) # output dataframe
for metric in metrics:
p_corr = get_fdr_p(df_pheno_z.loc[:, 'p'].filter(
regex=metric)) # correct p-values for metric
p_corr_tmp = pd.DataFrame(
index=df_pheno_z.loc[:, 'p'].filter(regex=metric).index,
columns=['p-corr'],
data=p_corr) # set to dataframe with correct indices
df_pheno_z.loc[p_corr_tmp.index,
'p-corr'] = p_corr_tmp # store using index matching
# In[23]:
for pheno in phenos:
for metric in metrics:
print(
pheno, metric,
np.sum(
df_pheno_z.filter(regex=metric, axis=0).filter(
#
# 1) Regions where there is a significant relationship between age and the regional brain features in the training set
#
# 2) Regions where the normative model was able to perform out of sample predictions (as index by standardized mean squared error < 1)
#
# 3) Regions where extreme deviations occur
# ### 1) Age effects
# In[26]:
# age effect on training set
df_age_effect = run_corr(df_train[primary_covariate],
df_node_train,
typ='pearsonr')
df_age_effect['p_fdr'] = get_fdr_p(df_age_effect['p'])
if parc_str == 'lausanne':
df_age_effect.drop(my_list, axis=0, inplace=True)
age_alpha = 0.05
age_filter = df_age_effect['p_fdr'].values < age_alpha
age_filter.sum()
# ### 2) Normative model performance
# In[27]:
smse_thresh = 1
smse_filter = df_smse.values < smse_thresh
smse_filter = smse_filter.reshape(-1)
smse_filter.sum()
nulldir=nulldir)
elif assign_p == 'parametric':
df_pheno = run_pheno_correlations(df.loc[:, phenos],
df_node,
method=method,
assign_p=assign_p)
# In[12]:
# correct multiple comparisons. We do this across brain regions and phenotypes (e.g., 400*6 = 2400 tests)
df_p_corr = pd.DataFrame(index=df_pheno.index,
columns=['p-corr']) # output dataframe
for metric in metrics:
p_corr = get_fdr_p(
df_pheno.loc[:,
'p'].filter(regex=metric)) # correct p-values for metric
p_corr_tmp = pd.DataFrame(
index=df_pheno.loc[:, 'p'].filter(regex=metric).index,
columns=['p-corr'],
data=p_corr) # set to dataframe with correct indices
df_pheno.loc[p_corr_tmp.index,
'p-corr'] = p_corr_tmp # store using index matching
# In[13]:
alpha = 0.05
print(alpha)
# In[14]: