def create_seq_kernels(self, all_pw_map, pat_ids, kms_file_name):
# experiment variables
num_pat = pat_ids.shape[0]
num_pw = len(all_pw_map)
kms_path = self.exp_data_dir / f'{kms_file_name}-{self.label}.npz'
if kms_path.exists(): return np_load_data(kms_path, key='kms')
# calculate kernel matrices for over expressed genes
over_exp_kms = np.zeros((num_pw, num_pat, num_pat))
for ind, (pw_id, pw) in enumerate(all_pw_map.items()): # for each pathway
over_exp_kms[ind] = kernel(pat_ids, pw, label_key=f'label-oe-{self.label}', alpha=self.smoothing_alpha,
normalization=self.kernel_normalization)
logr(f'Calculating oe pathway kernel {ind + 1:4}/{num_pat} pw_id={pw_id}')
log()
# calculate kernel matrices for under expressed genes
under_exp_kms = np.zeros((num_pw, num_pat, num_pat))
for ind, (pw_id, pw) in enumerate(all_pw_map.items()): # for each pathway
under_exp_kms[ind] = kernel(pat_ids, pw, label_key=f'label-ue-{self.label}', alpha=self.smoothing_alpha,
normalization=self.kernel_normalization)
logr(f'Calculating ue pathway kernel {ind + 1:4}/{num_pat} pw_id={pw_id}')
log()
kms = np.vstack([over_exp_kms, under_exp_kms]) # stack all kernels
np.savez_compressed(kms_path, kms=kms) # save kernels
return kms
def create_som_kernels(self, all_pw_map, patients):
# experiment variables
num_pat = len(patients)
num_pw = len(all_pw_map)
kms_path = self.exp_data_dir / 'som-kms.npz'
if kms_path.exists(): return np_load_data(kms_path, key='kms')
# calculate kernel matrices for over expressed genes
kms = np.zeros((num_pw, num_pat, num_pat))
pat_ids = np.array([pat['pat_id'] for pat in patients])
for ind, (pw_id, pw) in enumerate(all_pw_map.items()): # for each pathway
kms[ind] = kernel(pat_ids, pw, label_key='label-som', alpha=self.smoothing_alpha,
normalization=self.kernel_normalization)
logr(f'Calculating som mut pathway kernel {ind + 1:4}/{num_pat} pw_id={pw_id}')
log()
np.savez_compressed(kms_path, kms=kms) # save kernels
return kms
def create_cnv_kernels(self, all_pw_map, patients, cnv_type):
# experiment variables
num_pat = len(patients)
num_pw = len(all_pw_map)
kms_path = self.kernel_dir / f'cnv-kms-{cnv_type}.npz'
if kms_path.exists(): return np_load_data(kms_path, key='kms')
# calculate kernel matrices for over expressed genes
kms = np.zeros((num_pw, num_pat, num_pat))
pat_ids = np.array(list(patients.keys()))
for ind, (pw_id,
pw) in enumerate(all_pw_map.items()): # for each pathway
kms[ind] = kernel(pat_ids,
pw,
label_key=f'label-cnv_{cnv_type}',
alpha=self.smoothing_alpha,
normalization=self.kernel_normalization)
logr(
f'Calculating cnv {cnv_type} pathway kernel {ind + 1:4}/{num_pw} pw_id={pw_id}'
)
log()
np.savez_compressed(kms_path, kms=kms) # save kernels
return kms