def load_epimutations(config):
fn_dict = get_data_base_path(config) + '/' + 'epimutations_dict'
suffix = ''
if bool(config.experiment.data_params):
suffix += '_' + str(config.experiment.get_data_params_str())
fn_data = get_data_base_path(config) + '/' + 'epimutations' + suffix
if os.path.isfile(fn_dict + '.pkl') and os.path.isfile(fn_data + '.npz'):
f = open(fn_dict + '.pkl', 'rb')
config.epimutations_dict = pickle.load(f)
f.close()
data = np.load(fn_data + '.npz')
config.epimutations_data = data['data']
else:
config.experiment.data_params = {}
load_betas(config)
config.epimutations_dict = config.betas_dict
f = open(fn_dict + '.pkl', 'wb')
pickle.dump(config.epimutations_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
save_table_dict_csv(
fn_dict,
{
'item': list(config.epimutations_dict.keys()),
'row': list(config.epimutations_dict.values())
}
)
num_cpgs = config.betas_data.shape[0]
num_subjects = config.betas_data.shape[1]
config.epimutations_data = np.zeros((num_cpgs, num_subjects), dtype=np.int)
for cpg, row in tqdm(config.betas_dict.items(), mininterval=60.0):
betas = config.betas_data[row, :]
quartiles = np.percentile(betas, [25, 75])
iqr = quartiles[1] - quartiles[0]
left = quartiles[0] - (3.0 * iqr)
right = quartiles[1] + (3.0 * iqr)
curr_row = np.zeros(num_subjects, dtype=np.int)
for subject_id in range(0, num_subjects):
curr_point = betas[subject_id]
if curr_point < left or curr_point > right:
curr_row[subject_id] = 1
config.epimutations_data[row] = curr_row
np.savez_compressed(fn_data + '.npz', data=config.epimutations_data)
np.savetxt(fn_data + '.txt', config.epimutations_data, delimiter='\t', fmt='%d')
# Clear data
del config.betas_data
def load_entropy(config):
suffix = ''
if bool(config.experiment.data_params):
data_params = copy.deepcopy(config.experiment.data_params)
suffix += '_' + config.experiment.get_data_params_str()
else:
raise ValueError('Exog for entropy is empty.')
fn_data = get_cache_path(config) + '/' + 'entropy' + suffix + '.npz'
config.entropy_list = ['entropy']
config.entropy_dict = {'entropy': 0}
config.entropy_missed_dict = {'entropy': []}
if os.path.isfile(fn_data):
data = np.load(fn_data)
config.entropy_data = data['data']
else:
if data_params['data'] == 'betas':
config.experiment.data_params = {}
load_betas(config)
data = config.betas_data
data_dict = config.betas_dict
elif data_params['data'] == 'betas_adj':
config.experiment.data_params.pop('data')
load_betas_adj(config)
data = config.betas_adj_data
data_dict = config.betas_adj_dict
elif data_params['data'] == 'residuals':
config.experiment.data_params.pop('data')
load_residuals(config)
data = config.residuals_data
data_dict = config.residuals_dict
else:
raise ValueError('Unsupported data for entropy.')
num_subjects = data.shape[1]
config.entropy_data = np.zeros(num_subjects, dtype=np.float32)
rows = [data_dict[item] for item in config.cpg_list if item in data_dict]
for subj_id in tqdm(range(0, num_subjects), mininterval=60.0, desc='entropy_data creating'):
values = np.squeeze(np.asarray(data[np.ix_(rows, [subj_id])]))
entropy = 0.0
outliers = 0
for val in values:
if not math.isnan(val):
if 0.0 < val < 1.0:
entropy += val * np.log2(val) + (1.0 - val) * np.log2(1.0 - val)
else:
outliers += 1
else:
outliers += 1
entropy /= ((len(values) - outliers) * np.log2(0.5))
config.entropy_data[subj_id] = entropy
np.savez_compressed(fn_data, data=config.entropy_data)
def test_load_cpg_check_files_creation(self):
fn_dict = get_data_base_path(self.config) + '/' + 'betas_dict.pkl'
fn_data = get_data_base_path(self.config) + '/' + 'betas'
fn_npz = fn_data + '.npz'
load_betas(self.config)
self.assertEqual(True,
os.path.isfile(fn_dict) and os.path.isfile(fn_npz))
def load(self, config, configs_child):
load_betas(config)
config.base_list = config.cpg_list
config.base_dict = config.betas_adj_dict
config.base_data = config.betas_adj_data
self.inherit_childs(config, configs_child)
if config.is_load_child:
for config_child in configs_child:
self.load_child(config_child)
def load_betas_spec(config):
file = config.experiment.data_params.pop('file', None)
filename, file_extension = os.path.splitext(file)
fn_list = get_data_base_path(config) + '/' + file
fn_data_csv = get_data_base_path(config) + '/' + filename + '.csv'
if os.path.isfile(fn_list):
f = open(fn_list)
cpgs = f.readlines()
cpgs = [x.rstrip() for x in cpgs]
f.close()
else:
raise ValueError('File with cpgs is not exists.')
if not os.path.isfile(fn_data_csv):
indexes = config.attributes_indexes
fn_betas = get_data_base_path(config) + '/' + 'betas.txt'
f = open(fn_betas)
header_line = f.readline()
headers = get_line_list(header_line)
subject_headers = list(np.array(headers[1::])[indexes])
f.close()
load_betas(config)
num_cols = len(subject_headers) + 1 # header col
num_rows = len(cpgs) + 1 # header row
betas = np.zeros((num_rows, num_cols), dtype=object)
row_id = 0
betas[row_id] = ['ProbeID'] + subject_headers
row_id += 1
for cpg in tqdm(cpgs, mininterval=60.0, desc='betas_spec creating'):
if cpg in config.betas_dict:
cpg_row_id = config.betas_dict[cpg]
curr_betas = list(
np.array(config.betas_data[cpg_row_id])[indexes])
for missed_id in config.betas_missed_dict[cpg]:
curr_betas[missed_id] = 'NaN'
line = [cpg] + curr_betas
betas[row_id] = line
else:
line = [cpg] + ['NaN'] * (num_cols - 1)
betas[row_id] = line
row_id += 1
pd.DataFrame(betas).to_csv(fn_data_csv, index=False, header=False)
def load_betas_horvath_calculator(config):
fn_list = get_data_base_path(config) + '/' + 'cpgs_horvath_calculator.txt'
fn_data_csv = get_data_base_path(config) + '/' + 'betas_horvath_calculator.csv'
suffix = ''
if bool(config.experiment.data_params):
suffix += '_' + config.experiment.get_data_params_str()
if os.path.isfile(fn_list):
f = open(fn_list)
cpgs = f.readlines()
cpgs = [x.rstrip() for x in cpgs]
f.close()
else:
raise ValueError('No specified file with cpgs for Horvath\'s calculator.')
if not os.path.isfile(fn_data_csv):
fn_betas = get_data_base_path(config) + '/' + 'betas' + suffix + '.txt'
f = open(fn_betas)
header_line = f.readline()
headers = get_line_list(header_line)
load_betas(config)
num_cols = len(headers)
num_rows = len(cpgs) + 1 # header row
betas = np.zeros((num_rows, num_cols), dtype=object)
row_id = 0
betas[row_id] = ['ProbeID'] + headers[1::]
row_id += 1
for cpg in tqdm(cpgs, mininterval=60.0, desc='betas_horvath_calculator creating'):
if cpg in config.betas_dict:
cpg_row_id = config.betas_dict[cpg]
curr_betas = list(config.betas_data[cpg_row_id])
for missed_id in config.betas_missed_dict[cpg]:
curr_betas[missed_id] = 'NaN'
line = [cpg] + curr_betas
betas[row_id] = line
else:
line = [cpg] + ['NaN'] * (num_cols - 1)
betas[row_id] = line
row_id += 1
pd.DataFrame(betas).to_csv(fn_data_csv, index=False, header=False)
def load(self, config, configs_child):
if config.is_init:
source = config.experiment.data_params.pop('source')
if source == 'betas':
load_betas(config)
config.base_missed_dict = config.betas_missed_dict
config.base_data = config.betas_data
config.target_dict = config.betas_dict
elif source == 'residuals':
load_residuals(config)
config.base_missed_dict = config.residuals_missed_dict
config.base_data = config.residuals_data
config.target_dict = config.residuals_dict
config.base_list = list(config.bops.keys())
config.base_dict = config.bops
self.inherit_childs(config, configs_child)
if config.is_load_child:
for config_child in configs_child:
self.load_child(config_child)
def load_genes(config):
suffix_gene = ''
if bool(config.experiment.data_params):
suffix_gene += '_' + str(config.experiment.get_data_params_str())
source = config.experiment.data_params.pop('source')
else:
raise ValueError('Data params for genes are empty')
fn_list_txt = get_cache_path(config) + '/' + 'genes_list.txt'
fn_list_pkl = get_cache_path(config) + '/' + 'genes_list.pkl'
fn_dict_pkl = get_cache_path(config) + '/' + 'genes_dict.pkl'
fn_missed_dict_pkl = get_cache_path(config) + '/' + 'genes_missed_dict.pkl'
fn_data_npz = get_cache_path(config) + '/' + 'genes' + suffix_gene + '.npz'
fn_data_txt = get_cache_path(config) + '/' + 'genes' + suffix_gene + '.txt'
if os.path.isfile(fn_dict_pkl) and os.path.isfile(
fn_list_pkl) and os.path.isfile(fn_data_npz):
f = open(fn_list_pkl, 'rb')
config.genes_list = pickle.load(f)
f.close()
f = open(fn_dict_pkl, 'rb')
config.genes_dict = pickle.load(f)
f.close()
f = open(fn_missed_dict_pkl, 'rb')
config.genes_missed_dict = pickle.load(f)
f.close()
data = np.load(fn_data_npz)
config.genes_data = data['data']
else:
if source == 'betas':
load_betas(config)
source_dict = config.betas_dict
source_data = config.betas_data
source_missed_dict = config.betas_missed_dict
elif source == 'betas_adj':
load_betas_adj(config)
source_dict = config.betas_adj_dict
source_data = config.betas_adj_data
source_missed_dict = config.betas_adj_missed_dict
elif source == 'residuals':
load_residuals(config)
source_dict = config.residuals_dict
source_data = config.residuals_data
source_missed_dict = config.residuals_missed_dict
else:
raise ValueError('Source for genes is not specified')
num_subjects = config.betas_data.shape[1]
config.genes_list = []
for gene_id, gene in tqdm(enumerate(config.gene_cpg_dict),
mininterval=60.0,
desc='genes_list creating'):
cpgs = config.gene_cpg_dict[gene]
for cpg in cpgs:
if cpg in source_dict:
config.genes_list.append(gene)
break
config.genes_dict = {}
config.genes_missed_dict = {'any': []}
config.genes_data = np.zeros((len(config.genes_list), num_subjects),
dtype=np.float32)
for gene_id, gene in tqdm(enumerate(config.genes_list),
mininterval=60.0,
desc='genes_data creating'):
config.genes_dict[gene] = gene_id
cpgs = config.gene_cpg_dict[gene]
denominators = np.zeros(num_subjects, dtype=np.float32)
for cpg in cpgs:
if cpg in source_dict:
row_id = source_dict[cpg]
source_values_raw = source_data[row_id, :]
if len(source_missed_dict[cpg]) > 0:
source_values = np.zeros(num_subjects,
dtype=np.float32)
for val_id in range(0, num_subjects):
if val_id not in source_missed_dict[cpg]:
source_values[val_id] = source_values_raw[
val_id]
denominators[val_id] += 1.0
else:
source_values = source_values_raw
denominators += np.ones(num_subjects, dtype=np.float32)
config.genes_data[gene_id] += source_values
for val_id in range(0, num_subjects):
config.genes_data[gene_id][val_id] /= denominators[val_id]
f = open(fn_list_pkl, 'wb')
pickle.dump(config.genes_list, f, pickle.HIGHEST_PROTOCOL)
f.close()
f = open(fn_dict_pkl, 'wb')
pickle.dump(config.genes_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
f = open(fn_missed_dict_pkl, 'wb')
pickle.dump(config.genes_missed_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
np.savez_compressed(fn_data_npz, data=config.genes_data)
np.savetxt(fn_data_txt, config.genes_data, delimiter='\t', fmt='%.8e')
with open(fn_list_txt, 'w') as f:
for item in config.genes_list:
f.write("%s\n" % item)
def load_betas_adj(config):
suffix = ''
if bool(config.experiment.data_params):
suffix += '_' + config.experiment.get_data_params_str()
else:
raise ValueError('Exog for residuals is empty.')
fn_dict = get_data_base_path(
config) + '/' + 'betas_adj_dict' + suffix + '.pkl'
fn_missed_dict = get_data_base_path(
config) + '/' + 'betas_adj_missed_dict' + suffix + '.pkl'
fn_data = get_data_base_path(config) + '/' + 'betas_adj' + suffix + '.npz'
if os.path.isfile(fn_dict) and os.path.isfile(fn_data):
f = open(fn_dict, 'rb')
config.betas_adj_dict = pickle.load(f)
f.close()
f = open(fn_missed_dict, 'rb')
config.betas_adj_missed_dict = pickle.load(f)
f.close()
data = np.load(fn_data)
config.betas_adj_data = data['data']
else:
load_residuals(config)
config.experiment.data_params = {}
load_betas(config)
config.betas_adj_dict = config.residuals_dict
f = open(fn_dict, 'wb')
pickle.dump(config.betas_adj_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
config.betas_adj_missed_dict = config.residuals_missed_dict
f = open(fn_missed_dict, 'wb')
pickle.dump(config.betas_missed_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
num_cpgs = config.betas_data.shape[0]
num_subjects = config.betas_data.shape[1]
config.betas_adj_data = np.zeros((num_cpgs, num_subjects),
dtype=np.float32)
for cpg in tqdm(config.betas_adj_dict,
mininterval=60.0,
desc='betas_adj_data creating'):
residuals = config.residuals_data[config.residuals_dict[cpg], :]
betas = config.betas_data[config.betas_dict[cpg], :]
min_residuals = np.min(residuals)
mean_betas = np.mean(betas)
shift = mean_betas
if min_residuals + shift < 0:
shift = abs(min_residuals)
betas_adj = residuals + shift
config.betas_adj_data[config.residuals_dict[cpg]] = betas_adj
np.savez_compressed(fn_data, data=config.betas_adj_data)
# Clear data
del config.residuals_data
del config.betas_data
def load_betas_adj(config):
fn_dict = get_data_base_path(config) + '/' + 'betas_adj_dict.pkl'
suffix = ''
if bool(config.experiment.data_params):
data_params = config.experiment.data_params
suffix += '_' + config.experiment.get_data_params_str()
else:
raise ValueError(f'Exog for residuals is empty.')
fn_data = get_data_base_path(config) + '/' + 'betas_adj' + suffix + '.npz'
if os.path.isfile(fn_dict) and os.path.isfile(fn_data):
f = open(fn_dict, 'rb')
config.betas_adj_dict = pickle.load(f)
f.close()
data = np.load(fn_data)
config.betas_adj_data = data['data']
else:
config.experiment.data_params = {}
load_betas(config)
config.betas_adj_dict = config.betas_dict
f = open(fn_dict, 'wb')
pickle.dump(config.betas_adj_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
exog_dict = {}
if 'cells' in data_params:
cells_dict = load_cells_dict(config)
if isinstance(data_params['cells'], list):
all_types = list(cells_dict.keys())
for key in all_types:
if key not in data_params['cells']:
cells_dict.pop(key)
if len(list(cells_dict.keys())) != len(data_params['cells']):
raise ValueError(f'Wrong number of cells types.')
exog_dict.update(cells_dict)
if 'observables' in data_params:
observables_dict = load_observables_dict(config)
if isinstance(data_params['observables'], list):
all_types = list(observables_dict.keys())
for key in all_types:
if key not in data_params['observables']:
observables_dict.pop(key)
if len(list(observables_dict.keys())) != len(
data_params['observables']):
raise ValueError(f'Wrong number of observables types.')
exog_dict.update(observables_dict)
exog_df = pd.DataFrame(exog_dict)
num_cpgs = config.betas_data.shape[0]
num_subjects = config.betas_data.shape[1]
config.betas_adj_data = np.zeros((num_cpgs, num_subjects),
dtype=np.float32)
for cpg, row in tqdm(config.betas_dict.items(),
mininterval=60.0,
desc='betas_adj_data creating'):
betas = config.betas_data[row, :]
mean = np.mean(betas)
endog_dict = {cpg: betas}
endog_df = pd.DataFrame(endog_dict)
reg_res = sm.OLS(endog=endog_df, exog=exog_df).fit()
residuals = list(map(np.float32, reg_res.resid))
betas_adj = residuals + mean
config.betas_adj_data[row] = betas_adj
np.savez_compressed(fn_data, data=config.betas_adj_data)
# Clear data
del config.betas_data
def load_resid_old(config):
suffix = ''
if bool(config.experiment.data_params):
data_params = config.experiment.data_params
suffix += '_' + config.experiment.get_data_params_str()
else:
raise ValueError('Exog for resid_old is empty.')
fn_dict = get_data_base_path(
config) + '/' + 'resid_old_dict' + suffix + '.pkl'
fn_missed_dict = get_data_base_path(
config) + '/' + 'resid_old_missed_dict' + suffix + '.pkl'
fn_data = get_data_base_path(config) + '/' + 'resid_old' + suffix + '.npz'
if os.path.isfile(fn_dict) and os.path.isfile(fn_data):
f = open(fn_dict, 'rb')
config.resid_old_dict = pickle.load(f)
f.close()
f = open(fn_missed_dict, 'rb')
config.resid_old_missed_dict = pickle.load(f)
f.close()
data = np.load(fn_data)
config.resid_old_data = data['data']
else:
data_params_copy = copy.deepcopy(config.experiment.data_params)
common_keys = ['norm']
config.experiment.data_params = {}
for key in common_keys:
if key in data_params_copy:
config.experiment.data_params[key] = data_params_copy[key]
load_betas(config)
config.resid_old_dict = config.betas_dict
f = open(fn_dict, 'wb')
pickle.dump(config.resid_old_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
config.resid_old_missed_dict = config.betas_missed_dict
f = open(fn_missed_dict, 'wb')
pickle.dump(config.resid_old_missed_dict, f, pickle.HIGHEST_PROTOCOL)
f.close()
exog_dict = {}
if 'cells' in data_params:
cells_dict = load_cells_dict(config)
if isinstance(data_params['cells'], list):
all_types = list(cells_dict.keys())
for key in all_types:
if key not in data_params['cells']:
cells_dict.pop(key)
if len(list(cells_dict.keys())) != len(data_params['cells']):
raise ValueError('Wrong number of cells types.')
exog_dict.update(cells_dict)
if 'observables' in data_params:
observables_categorical_dict = load_observables_categorical_dict(
config)
if isinstance(data_params['observables'], list):
all_types = list(observables_categorical_dict.keys())
for key in all_types:
if key not in data_params['observables']:
observables_categorical_dict.pop(key)
if len(list(observables_categorical_dict.keys())) != len(
data_params['observables']):
raise ValueError('Wrong number of observables types.')
exog_dict.update(observables_categorical_dict)
num_cpgs = config.betas_data.shape[0]
num_subjects = config.betas_data.shape[1]
config.resid_old_data = np.zeros((num_cpgs, num_subjects),
dtype=np.float32)
for cpg, row in tqdm(config.betas_dict.items(),
mininterval=60.0,
desc='resid_old_data creating'):
raw_betas = config.betas_data[row, :]
current_exog_dict = copy.deepcopy(exog_dict)
if len(config.betas_missed_dict[cpg]) > 0:
for key in current_exog_dict:
values = []
for value_id in range(0, len(current_exog_dict[key])):
if value_id not in config.betas_missed_dict[cpg]:
values.append(current_exog_dict[key][value_id])
current_exog_dict[key] = values
betas = []
passed_ids = []
for beta_id in range(0, len(raw_betas)):
if beta_id not in config.betas_missed_dict[cpg]:
betas.append(raw_betas[beta_id])
passed_ids.append(beta_id)
else:
betas = raw_betas
passed_ids = list(range(0, len(betas)))
endog_dict = {cpg: betas}
endog_df = pd.DataFrame(endog_dict)
exog_df = pd.DataFrame(current_exog_dict)
reg_res = sm.OLS(endog=endog_df, exog=exog_df).fit()
resid_old = list(map(np.float32, reg_res.resid))
resid_old_raw = np.zeros(num_subjects, dtype=np.float32)
for beta_id in range(0, len(passed_ids)):
resid_old_raw[passed_ids[beta_id]] = resid_old[beta_id]
for missed_id in config.resid_old_missed_dict[cpg]:
resid_old_raw[missed_id] = np.float32('nan')
config.resid_old_data[row] = resid_old_raw
np.savez_compressed(fn_data, data=config.resid_old_data)
# Clear data
del config.betas_data
def test_load_cpg_check_shape_cpg_data(self):
load_betas(self.config)
self.assertEqual((300, 729), self.config.betas_data.shape)
def test_load_cpg_check_len_cpg_dict(self):
load_betas(self.config)
self.assertEqual(300, len(list(self.config.betas_dict)))
def load_epimutations(config):
suffix = ''
if bool(config.experiment.data_params):
suffix += '_' + str(config.experiment.get_data_params_str())
fn_data = get_data_base_path(config) + '/' + 'epimutations' + suffix
config.epimutations_list = ['epimutations']
config.epimutations_dict = {'epimutations': 0}
config.epimutations_missed_dict = {'epimutations': []}
data_params_copy = copy.deepcopy(config.experiment.data_params)
common_keys = ['part', 'norm']
config.experiment.data_params = {}
for key in common_keys:
if key in data_params_copy:
config.experiment.data_params[key] = data_params_copy[key]
load_betas(config)
if os.path.isfile(fn_data + '.npz'):
data = np.load(fn_data + '.npz')
config.epimutations_data = data['data']
else:
num_cpgs = config.betas_data.shape[0]
num_subjects = config.betas_data.shape[1]
config.epimutations_data = np.zeros((num_cpgs, num_subjects),
dtype=np.int)
for cpg, row in tqdm(config.betas_dict.items(),
mininterval=60.0,
desc='epimutations_data creating'):
betas_raw = config.betas_data[row, :]
if len(config.betas_missed_dict[cpg]) > 0:
betas = []
for beta_id in range(0, len(betas_raw)):
if beta_id not in config.betas_missed_dict[cpg]:
betas.append(betas_raw[beta_id])
else:
betas = betas_raw
quartiles = np.percentile(betas, [25, 75])
iqr = quartiles[1] - quartiles[0]
left = quartiles[0] - (3.0 * iqr)
right = quartiles[1] + (3.0 * iqr)
curr_row = np.zeros(num_subjects, dtype=np.int)
for subject_id in range(0, num_subjects):
curr_point = betas[subject_id]
if not math.isnan(curr_point):
if curr_point < left or curr_point > right:
curr_row[subject_id] = 1
config.epimutations_data[row] = curr_row
np.savez_compressed(fn_data + '.npz', data=config.epimutations_data)
np.savetxt(fn_data + '.txt',
config.epimutations_data,
delimiter='\t',
fmt='%d')
# Clear data
del config.betas_data
