def do_test(self, seed, fasta_file):
"Do one test case."
#
# Load sequences and build index
#
algorithm = stempy.Algorithm(self.options)
algorithm._initialise(fasta_file)
data = algorithm.input_sequences.data
#
# look for best W-mers
#
W = len(seed)
num_to_find = 3
logging.info('Looking for %d best W-mers', num_to_find)
best_w_mer_finder = stempy.create_best_w_mer_finder(
data,
algorithm.create_model_of_input(W),
num_to_find
)
logging.info('Seeding model with %s', seed)
if W != len(seed):
raise ValueError('Seed must be same length as motif.')
best_w_mer_finder.model.bs.seed(seed, True)
best_w_mer_finder.model.set_lambda_for_sites(best_w_mer_finder.data.num_sequences)
best_w_mer_finder()
if not best_w_mer_finder.best_w_mers:
raise RuntimeError('Did not find any W-mers')
#
# Log best W-mers
#
for _eval in best_w_mer_finder.best_w_mers:
seq, offset = data.pos_localise(_eval.global_pos)
strand = _eval.rev_comp and '-' or '+'
logging.info(
'Seed: %s; Site: %s; seq: % 2d; offset: % 4d; strand: %s; p(binding): %.2e; p(not binding): %.2e',
seed, data.get_W_mer(W, _eval.global_pos), seq, offset, strand, _eval.Z, 1.-_eval.Z
)
#
# check we at least found the seed...
#
for _eval in best_w_mer_finder.best_w_mers:
if data.get_W_mer(W, _eval.global_pos) == seed:
break
else:
raise RuntimeError('Could not find seed in best W-mers')
#
# check we have no overlaps
#
localised_positions = [data.pos_localise(_eval.global_pos) for _eval in best_w_mer_finder.best_w_mers]
check_w_mers_dont_overlap(W, localised_positions)
assert not best_w_mer_finder.has_overlapping
def test_find_best_W_mers_2(self):
"""
Run best W-mer finder and check we found the seeds we wanted.
"""
self.options.min_num_sites = self.options.max_num_sites = num_to_find = 2
# load data and create STEME object
fasta_file = os.path.normpath(get_fasta_file('T00759-small.fa'))
#
# Load sequences and build index
#
algorithm = stempy.Algorithm(self.options)
algorithm._initialise(fasta_file)
data = algorithm.input_sequences.data
for seed in (
'ATGCAGAAAAATTAAG',
'TTTAAAATACTTTAAA',
):
# create and seed a model
W = len(seed)
model = algorithm.create_model_of_input(W)
model.bs.seed(seed, True)
model.set_lambda_for_sites(data.num_sequences)
# look for best W-mers under model
best_w_mer_finder = stempy.create_best_w_mer_finder(data, model, num_to_find)
best_w_mer_finder()
avg_Z = 0.
for _eval in best_w_mer_finder.best_w_mers:
logging.info(
'Seed: %s; Site: %s; p(binding): %.2e; p(not binding): %.2e',
seed, data.get_W_mer(W, _eval.global_pos), _eval.Z, 1.-_eval.Z
)
avg_Z += _eval.Z
logging.info('Seed: %s; Average Z: %.6f', seed, avg_Z / len(best_w_mer_finder.best_w_mers))
#
# Check we found the seed
#
for _eval in best_w_mer_finder.best_w_mers:
if data.get_W_mer(W, _eval.global_pos) == seed:
break
else:
raise RuntimeError('Could not find seed in best W-mers')
#
# Log the product of p-values
#
best_w_mer_finder.update_model(num_to_find, use_pseudo_counts=False)
logging.info('Seed: %s; log PoP: %.6f', seed, algorithm.significance.log_product_p_values(model))
def test_specific_case(self):
"""
Check we get the correct W-mers.
"""
fasta_file = os.path.normpath(get_fasta_file('T00759-tiny.fa'))
seed = 'AAAACCCA'
W = len(seed)
num_sites = 4
self.options.max_num_sites = num_sites
self.options.min_num_sites = num_sites
#
# Load sequences and build index
#
algorithm = stempy.Algorithm(self.options)
algorithm._initialise(fasta_file)
data = algorithm.input_sequences.data
model = algorithm.create_model_of_input(W)
model.bs.seed(seed, True)
model.set_lambda_for_sites(data.num_sequences)
# look for best W-mers under model
best_w_mer_finder = stempy.create_best_w_mer_finder(data, model, num_sites)
best_w_mer_finder()
if len(best_w_mer_finder.best_w_mers) < num_sites:
if len(best_w_mer_finder.best_w_mers) != model.num_W_mers:
raise ValueError('Did not find enough W-mers')
# We want to get these W-mers
#
# 2011-08-09 10:11:32,846 - INFO - Z=8.00e-02; pos= 313 +; AAAACCCA; AAAACCCA
# 2011-08-09 10:11:32,846 - INFO - Z=4.37e-02; pos= 668 -; TGAGTTTT; AAAACTCA
# 2011-08-09 10:11:32,846 - INFO - Z=1.37e-02; pos= 710 -; TGGTTTTC; GAAAACCA
# 2011-08-09 10:11:32,846 - INFO - Z=1.37e-02; pos= 681 -; TGGTTCTT; AAGAACCA
#
for wmer, (global_pos, rev_comp) in zip(best_w_mer_finder.best_w_mers, [(313, False), (668, True), (710, True), (681, True)]):
if wmer.global_pos != global_pos and wmer.Z < model.calculate_Z(global_pos, rev_comp):
raise ValueError('Got wrong W-mers')
def find_best_w_mers_for_seed(stem_algorithm, seed, options, log_level):
"Find the best W-mers for the seed."
num_to_find = options.max_num_sites
logging.log(
log_level, 'Finding best %d W-mers for seed %s', num_to_find, seed)
_freqs, _freqs_with_pseudo_counts, model = stempy.Model(
stem_algorithm.data,
len(seed),
options
)
logging.debug('Seed pseudo-count: %f', model.bs.seed_pseudo_counts)
logging.debug('Model lambda: %f', model.lambda_)
model.bs.seed(seed, True)
best_w_mer_finder = stempy.create_best_w_mer_finder(
stem_algorithm.data, model, num_to_find)
best_w_mer_finder()
best_w_mers = list(best_w_mer_finder.best_w_mers)[-num_to_find:]
avg_Z = 0.
avg_distance = 0.
for eval_ in best_w_mers:
w_mer = stem_algorithm.data.get_W_mer(len(seed), eval_.global_pos)
if eval_.rev_comp:
w_mer = reverse_complement(w_mer)
distance = hamming_distance(seed, w_mer)
logging.log(log_level, 'Best W-mer: %s; Z=%.5f, distance=%d',
w_mer, eval_.Z, distance)
avg_Z += eval_.Z
avg_distance += distance
best_w_mer_finder.update_model(num_to_find, use_pseudo_counts=False)
log_pop = model.log_product_of_pvalues
avg_Z /= len(best_w_mers)
logging.log(log_level, 'Seed: %s; log PoP: %.6f', seed, log_pop)
logging.log(log_level, 'Seed: %s; Average Z=%.7f', seed, avg_Z)
logging.log(log_level, 'Seed: %s; Average distance=%.2f',
seed, avg_distance / len(best_w_mers))
logging.debug('Frequencies: %s' %
' '.join(map(str, (model.bg.freqs.freq(x) for x in xrange(4)))))
return avg_Z, log_pop
def test_we_get_all_W_mers_we_asked_for(self):
"""
Check that we are not short-changed on the number of W-mers we asked for.
"""
fasta_file = os.path.normpath(get_fasta_file('T00759-small.fa'))
num_sites = [2, 4, 8, 16, 32]
self.options.max_num_sites = max(num_sites)
self.options.min_num_sites = min(num_sites)
#
# Load sequences and build index
#
algorithm = stempy.Algorithm(self.options)
algorithm._initialise(fasta_file)
data = algorithm.input_sequences.data
for seed in (
'GCTAGCTAGCGG',
'ATGCAGAAAAATTAAG',
'TTTAAAATACTTTAAA',
):
# seed a model
logging.info('Using seed %s', seed)
W = len(seed)
model = algorithm.create_model_of_input(W)
model.bs.seed(seed, True)
model.set_lambda_for_sites(data.num_sequences)
for num_to_find in num_sites:
# look for best W-mers under model
best_w_mer_finder = stempy.create_best_w_mer_finder(data, model, num_to_find)
best_w_mer_finder()
if len(best_w_mer_finder.best_w_mers) < num_to_find:
if len(best_w_mer_finder.best_w_mers) != model.num_W_mers:
logging.warning('Found %d W-mers', len(best_w_mer_finder.best_w_mers))
logging.warning('%d W-mers available', model.num_W_mers)
logging.warning('Wanted %d W-mers', num_to_find)
raise ValueError('Did not find enough W-mers')
bs_model = stempy.PssmBindingSiteModel(stempy.initialise_uniform_pssm(W, options.alphabet_size))
bs_model.seed(seed)
# whole model
model = stempy.Model(data, bs_model, bg_model, _lambda=0.)
Z_threshold = .3
with Timer(msg='find instances with Z>%f' % Z_threshold):
instance_finder = stempy.FindInstances(data, model, Z_threshold)
instance_finder()
logging.info('Found %d instances', len(instance_finder.instances))
num_W_mers_to_find = 10000
with Timer(msg='find %d best W-mers' % num_W_mers_to_find):
w_mer_finder = stempy.create_best_w_mer_finder(data, model, num_W_mers_to_find)
w_mer_finder()
logging.info('Found %d instances', len(w_mer_finder.best_w_mers))
def global_overlap(pos1, pos2, W):
return abs(pos1 - pos2) < W
def get_non_overlapping(instances, W):
instances.sort()
instances.reverse()
result = []
for i in instances:
for better in result:
if global_overlap(i.global_pos, better.global_pos, W):
break