def _validate_fs(self, feature_selection):
if feature_selection not in self.FEATURE_SELECTION:
common.terminate(
"Choose feature selection from the options: %s (selected fs: %s)."
% (self.FEATURE_SELECTION, feature_selection))
if self.use_phenos is None:
if feature_selection == 'phenotype':
common.terminate(
"Must provide a phenotype (--pheno) when selecting feature selection of type 'phenotype'."
)
elif feature_selection == 'controls':
common.terminate(
"Must provide one phenotype (--pheno) in a binary format when selection feature selection of type 'controls'."
)
return getattr(
self,
self.FEATURE_FUNC_NAME_FORMAT.format(
feature_option_name=feature_selection))
def validate_args(self, args):
# make sure user choose only one EWAS test (mutually exlusive group is not supported...)
# argument pheno is required for all ewas tests - it can be supplied through --pheno flag of .glint meth data file
# So, if the datafile supplied is not .glint file - pheno must be supplied as a flag
if not args.datafile.name.endswith(methylation_data.GLINT_FILE_SUFFIX):
self.required_args.append('phenofile')
super(EWASParser, self).validate_args(args)
if len(args.pheno
) > 1: # 0 is all phenotypes in the data (which could be one)
common.terminate("Must supply only one phenotype for EWAS")
test_counter = 0
if args.lmm:
test_counter += 1
if args.logreg:
test_counter += 1
if args.linreg:
test_counter += 1
if args.wilc:
if args.covar:
common.terminate(
"Wilcoxon test cannot take any covaraites. Remove the --covar argument."
)
test_counter += 1
if test_counter > 1:
common.terminate("Select only one EWAS test.")
# add lmm parser if lmm was chosen
if args.lmm:
self.lmm_parser.validate_args(args)
self.all_args.extend(self.lmm_parser.all_args)
self.required_args.extend(self.lmm_parser.required_args)
# default test is linear regression
if test_counter == 0:
args.linreg = True
logging.info(
"No EWAS test was chosen, using the default linear regression."
)
def impute(self, min_score, plink_snp_file, plink_geno_file,
plink_ind_file, min_missing_values):
"""
impute with following rules:
replace missing values with mean (unless there are more htan min_missing_values missing values)
remove samples (dont impute them) which have more than min_missing_values missing snps (out of all its snps)
remove snps (sont impute with them) which have more than min_missing_values missing samples (out of all its samples)
remore site with score lower than min_score
dont impute sites that we dont have any snp (if we have at least one - impute it)
"""
samples = loadtxt(plink_ind_file, dtype=str, usecols=(0, ))
number_of_samples = samples.shape[0]
filename = plink_ind_file.name if type(
plink_ind_file) == file else plink_ind_file
logging.info("Found %s samples in the file '%s'." %
(number_of_samples, filename))
if type(plink_ind_file) == file:
plink_ind_file.close()
plink_snps_data = loadtxt(plink_snp_file, dtype=str)
if type(plink_snp_file) == file:
plink_snp_file.close()
# use only snps that their allele are not the pairs CG or AT - since in those cases we cannot know which strand was tested
relevant_snps_indices = self.get_relevant_plink_snp_list(
plink_snps_data)
logging.info("Get number of snp occurences...")
# extract find occurences per sample for each snp from .geno file
# snp_occurrences, relevant_snps_indices, missing_sampels_indices, non_missing_sampels_indices = self.get_snps_occurences(plink_geno_file, relevant_snps_indices, number_of_samples, min_missing_values) #missing samples handling
snp_occurrences, relevant_snps_indices = self.get_snps_occurences(
plink_geno_file, relevant_snps_indices, number_of_samples,
min_missing_values)
# logging.debug("samples removed %s" % ", ".join(samples[missing_sampels_indices])) #missing samples handling
# indices = [i for i,name in enumerate(plink_snps_data[relevant_snps_indices,0]) if name in self.snps_id_per_name]
relevant_snps_names = []
relevant_snp_occurrences = []
for i, name in enumerate(plink_snps_data[relevant_snps_indices, 0]):
if name in self.snps_id_per_name:
relevant_snps_names.append(name)
relevant_snp_occurrences.append(snp_occurrences[i])
# remove snps that we dont have information on
self.imputed_samples = samples #[non_missing_sampels_indices] #missing samples handling
number_of_samples = self.imputed_samples.size
if (number_of_samples == 0):
common.terminate(
"All samples were removed. There is nothing to impute. Quiting..."
)
# find sites with score bigger than min_score
seterr(
invalid='ignore'
) # to ignore the following line warning (These warnings are an intentional aspect of numpy)
relevant_sites_indices = where(self.sites_scores > min_score)[0]
# this code removes the sites in the list bad_sites_list (list of cpgs) from our data. it wast tested.
# logging.info("remove bad sites..." )
# assert type(bad_sites_list) == list or type(bad_sites_list) == ndarray
# assert type(self.sites_name_per_id) == list or type(self.sites_name_per_id) == ndarray
# bad_sites_indices = where(in1d(self.sites_name_per_id, bad_sites_list))[0]
# relevant_sites_indices = delete(relevant_sites_indices, bad_sites_indices)
# impute
logging.info("Impute methylation levels...")
site_imputation, imputed_sites_ids = self.impute_sites(
number_of_samples, relevant_snps_names, relevant_snp_occurrences,
relevant_sites_indices)
if site_imputation == []: # no sites imputed
common.terminate(
"All sites were removed. There is nothing to impute.")
logging.info("%s sites were imputed for %s samples." %
(len(imputed_sites_ids), number_of_samples))
logging.info(
"%s sites with score > %s were not imputed due to missing SNPs." %
(len(relevant_sites_indices) - len(imputed_sites_ids), min_score))
self.imputed_sites_names = self.sites_name_per_id[imputed_sites_ids]
self.site_imputation = site_imputation
def _validate_stdth(self, stdth):
if stdth > 1 or stdth < 0:
common.terminate(
"stdth cannot be greater than 1 and lower than 0. stdth = %s."
% stdth)
return stdth