def test_from_names(my_genes):
genome = ExpGenome.from_gene_names(my_genes)
assert len(genome) == len(my_genes)
for i, (g, eg) in enumerate(zip(my_genes, genome)):
assert eg.name == g
assert eg.chromosomes == []
assert eg.ensembl_ids == []
def my_genome():
# we're creating a fake genome consisting of all lowercase ascii characters
genes = [text(c) for c in ascii_lowercase]
genome = ExpGenome.from_gene_names(genes)
return genome
def run(self):
"""Perform GO-PCA.
Parameters
----------
Returns
-------
`GOPCARun` or None
The GO-PCA run, or ``None`` if the run failed.
"""
t0 = time.time() # remember the start time
timestamp = str(datetime.datetime.utcnow()) # timestamp for the run
### Phase 1: Make sure all configurations are valid
all_configs_valid = True
for config in self.configs:
if not config.user_params.check_params():
# problems with the configuration
all_configs_valid = False
config.finalize_params(self.matrix.p)
if not config.params.check_params():
all_configs_valid = False
if not all_configs_valid:
logger.error('Invalid configuration settings. '
'Aborting GO-PCA run.')
return None
# print some information
p, n = self.matrix.shape
logger.info('Timestamp: %s', timestamp)
logger.info(
'Size of expression matrix: ' + 'p=%d genes x n=%d samples.', p, n)
# Report hash values for expression matrix and configurations
expression_hash = self.matrix.hash
logger.info('Expression matrix hash: %s', expression_hash)
config_hashes = []
for i, config in enumerate(self.configs):
config_hashes.append(config.hash)
logger.info('Configuration #%d hash: %s', i + 1, config_hashes[-1])
### Phase 2: Determine the number of principal components
num_components = self.num_components
if num_components == 0:
# estimate the number of non-trivial PCs using a permutation test
num_components = self.estimate_num_components()
if num_components == 0:
logger.error('The estimated number of non-trivial '
'principal components is 0. '
'Aborting GO-PCA run.')
return None
if 0 < self.pc_max_components < num_components:
num_components = self.pc_max_components
logger.info('Limiting the number of PCs to test to %d.',
num_components)
else:
# determine the total number of principal components
# (i.e., the number of dimensions spanned by the data)
max_components = min(self.matrix.p, self.matrix.n - 1)
if self.num_components > max_components:
logger.error(
'The number of PCs to test was specified as '
'%d, but the data spans only %d dimensions. '
'Aborting GO-PCA run.', num_components, max_components)
return None
if num_components == 0:
logger.error('No principal components to test.'
'Aborting GO-PCA run.')
return None
### Phase 3: Perform PCA
logger.info('Performing PCA...')
pca = PCA(n_components=num_components)
Y = pca.fit_transform(self.matrix.X.T)
# output fraction of variance explained for the PCs tested
frac = pca.explained_variance_ratio_
cum_frac = np.cumsum(frac)
logger.info(
'Fraction of total variance explained by the first '
'%d PCs: %.1f%%', num_components, 100 * cum_frac[-1])
### Phase 4: Run GO-PCA for each configuration supplied
enr_logger = logging.getLogger(enrichment.__name__)
genome = ExpGenome.from_gene_names(self.matrix.genes.tolist())
W = pca.components_.T # the loadings matrix
msg = logger.debug
if self.verbose:
# enable more verbose "INFO" messages
msg = logger.info
all_signatures = []
for k, config in enumerate(self.configs):
logger.info(
'Generating GO-PCA signatures for configuration '
'%d...', k + 1)
# create GeneSetEnrichmentAnalysis object
enr_logger.setLevel(logging.ERROR)
gse_analysis = GeneSetEnrichmentAnalysis(genome, config.gene_sets)
enr_logger.setLevel(logging.NOTSET)
# generate signatures
final_signatures = []
var_expl = 0.0
for d in range(num_components):
var_expl += frac[d]
msg('')
msg('-' * 70)
msg('PC %d explains %.1f%% of the variance.', d + 1,
100 * frac[d])
msg(
'The new cumulative fraction of variance explained '
'is %.1f%%.', 100 * var_expl)
signatures_dsc = self._generate_pc_signatures(
self.matrix, config.params, gse_analysis, W, d + 1)
signatures_asc = self._generate_pc_signatures(
self.matrix, config.params, gse_analysis, W, -(d + 1))
signatures = signatures_dsc + signatures_asc
msg('# signatures: %d', len(signatures))
# apply global filter (if enabled)
if not config.params.no_global_filter:
before = len(signatures)
signatures = self._global_filter(config.params, signatures,
final_signatures,
config.gene_ontology)
msg('Global filter: kept %d / %d signatures.',
len(signatures), before)
# self.print_signatures(signatures, debug=True)
final_signatures.extend(signatures)
msg('Total no. of signatures generated so far: %d',
len(final_signatures))
logger.info('')
logger.info('=' * 70)
logger.info(
'GO-PCA for configuration #%d generated %d '
'signatures.', k + 1, len(final_signatures))
logger.info('-' * 70)
self.print_signatures(final_signatures)
logger.info('=' * 70)
logger.info('')
all_signatures.extend(final_signatures)
### Phase 5: Generate signature matrix and return a `GOPCARun` instance
sig_matrix = GOPCASignatureMatrix.from_signatures(all_signatures)
t1 = time.time()
exec_time = t1 - t0
logger.info('This GO-PCA run took %.2f s.', exec_time)
gopca_run = GOPCARun(sig_matrix, gopca.__version__, timestamp,
exec_time, expression_hash, config_hashes,
self.matrix.genes, self.matrix.samples, W, Y)
return gopca_run
def run(self):
"""Perform GO-PCA.
Parameters
----------
Returns
-------
`GOPCARun` or None
The GO-PCA run, or ``None`` if the run failed.
"""
t0 = time.time() # remember the start time
timestamp = str(datetime.datetime.utcnow()) # timestamp for the run
### Phase 1: Make sure all configurations are valid
all_configs_valid = True
for config in self.configs:
if not config.user_params.check_params():
# problems with the configuration
all_configs_valid = False
config.finalize_params(self.matrix.p)
if not config.params.check_params():
all_configs_valid = False
if not all_configs_valid:
logger.error('Invalid configuration settings. '
'Aborting GO-PCA run.')
return None
# print some information
p, n = self.matrix.shape
logger.info('Timestamp: %s', timestamp)
logger.info('Size of expression matrix: ' +
'p=%d genes x n=%d samples.', p, n)
# Report hash values for expression matrix and configurations
expression_hash = self.matrix.hash
logger.info('Expression matrix hash: %s', expression_hash)
config_hashes = []
for i, config in enumerate(self.configs):
config_hashes.append(config.hash)
logger.info('Configuration #%d hash: %s', i+1, config_hashes[-1])
### Phase 2: Determine the number of principal components
num_components = self.num_components
if num_components == 0:
# estimate the number of non-trivial PCs using a permutation test
num_components = self.estimate_num_components()
if num_components == 0:
logger.error('The estimated number of non-trivial '
'principal components is 0. '
'Aborting GO-PCA run.')
return None
if 0 < self.pc_max_components < num_components:
num_components = self.pc_max_components
logger.info('Limiting the number of PCs to test to %d.', num_components)
else:
# determine the total number of principal components
# (i.e., the number of dimensions spanned by the data)
max_components = min(self.matrix.p, self.matrix.n - 1)
if self.num_components > max_components:
logger.error('The number of PCs to test was specified as '
'%d, but the data spans only %d dimensions. '
'Aborting GO-PCA run.',
num_components, max_components)
return None
if num_components == 0:
logger.error('No principal components to test.'
'Aborting GO-PCA run.')
return None
### Phase 3: Perform PCA
logger.info('Performing PCA...')
pca = PCA(n_components=num_components)
Y = pca.fit_transform(self.matrix.X.T)
# output fraction of variance explained for the PCs tested
frac = pca.explained_variance_ratio_
cum_frac = np.cumsum(frac)
logger.info('Fraction of total variance explained by the first '
'%d PCs: %.1f%%', num_components, 100 * cum_frac[-1])
### Phase 4: Run GO-PCA for each configuration supplied
enr_logger = logging.getLogger(enrichment.__name__)
genome = ExpGenome.from_gene_names(self.matrix.genes.tolist())
W = pca.components_.T # the loadings matrix
msg = logger.debug
if self.verbose:
# enable more verbose "INFO" messages
msg = logger.info
all_signatures = []
for k, config in enumerate(self.configs):
logger.info('Generating GO-PCA signatures for configuration '
'%d...', k+1)
# create GeneSetEnrichmentAnalysis object
enr_logger.setLevel(logging.ERROR)
gse_analysis = GeneSetEnrichmentAnalysis(genome, config.gene_sets)
enr_logger.setLevel(logging.NOTSET)
# generate signatures
final_signatures = []
var_expl = 0.0
for d in range(num_components):
var_expl += frac[d]
msg('')
msg('-'*70)
msg('PC %d explains %.1f%% of the variance.',
d+1, 100*frac[d])
msg('The new cumulative fraction of variance explained '
'is %.1f%%.', 100*var_expl)
signatures_dsc = self._generate_pc_signatures(
self.matrix, config.params, gse_analysis, W, d+1)
signatures_asc = self._generate_pc_signatures(
self.matrix, config.params, gse_analysis, W, -(d+1))
signatures = signatures_dsc + signatures_asc
msg('# signatures: %d', len(signatures))
# apply global filter (if enabled)
if not config.params.no_global_filter:
before = len(signatures)
signatures = self._global_filter(
config.params, signatures, final_signatures,
config.gene_ontology)
msg('Global filter: kept %d / %d signatures.',
len(signatures), before)
# self.print_signatures(signatures, debug=True)
final_signatures.extend(signatures)
msg('Total no. of signatures generated so far: %d',
len(final_signatures))
logger.info('')
logger.info('='*70)
logger.info('GO-PCA for configuration #%d generated %d '
'signatures.', k+1, len(final_signatures))
logger.info('-'*70)
self.print_signatures(final_signatures)
logger.info('='*70)
logger.info('')
all_signatures.extend(final_signatures)
### Phase 5: Generate signature matrix and return a `GOPCARun` instance
sig_matrix = GOPCASignatureMatrix.from_signatures(all_signatures)
t1 = time.time()
exec_time = t1 - t0
logger.info('This GO-PCA run took %.2f s.', exec_time)
gopca_run = GOPCARun(sig_matrix,
gopca.__version__, timestamp, exec_time,
expression_hash, config_hashes,
self.matrix.genes, self.matrix.samples, W, Y)
return gopca_run
