y = np.vstack(zip(*label_y_train))
y_prob = np.vstack(zip(*label_pred_proba_train))
training_stats_i_f.merge(multi_label_evaluate(y, y_prob, threshold))
y = np.vstack(zip(*label_y_valid))
y_prob = np.vstack(zip(*label_pred_proba_valid))
validation_stats_i_f.merge(multi_label_evaluate(y, y_prob, threshold))
y = np.vstack(zip(*label_y_test))
y_prob = np.vstack(zip(*label_pred_proba_test))
testing_stats_i_f.merge(multi_label_evaluate(y, y_prob, threshold))
return training_stats_i_f, validation_stats_i_f, testing_stats_i_f
# For each iteration, batch the folds into parallel jobs
statistics_objects_i = parallel_map(do_fold, range(cv_folds), n_jobs)
for (train, val, test) in statistics_objects_i:
training_stats_i.merge(train)
validation_stats_i.merge(val)
testing_stats_i.merge(test)
log.write('Iteration {}\n'.format(i))
log.write('Training {}\n'.format(i))
training_stats_i.write(log, 'a')
log.write('Validation {}\n'.format(i))
validation_stats_i.write(log, 'a')
log.write('Testing {}\n'.format(i))
testing_stats_i.write(log, 'a')
statistics_objects.append([training_stats_i, validation_stats_i, testing_stats_i])
# return training_stats_i, validation_stats_i, testing_stats_i
p,
','.join(data[UniProt.GO]),
','.join(data[UniProt.IPR]),
','.join(data[UniProt.PFAM]),
','.join(data[UniProt.GO_EVD])
)
)
fp.close()
# ----------------------------- RUN SEMANTIC SIM IF NEEDED ---------------------------- #
if update or (not os.path.isfile(SIMILARITY_MAP)):
print("Computing semantic similarities...")
all_container = interactome.merge(train).merge(test)
all_ppis = all_container.remove_symmetric().get_ppis()
chunk_size = len(all_ppis) // n_jobs
sem_sim_tuples = parallel_map(compute_ss, chunks(all_ppis, chunk_size), n_jobs=n_jobs)
sem_sim_tuples = reduce(lambda x, y: x + y, sem_sim_tuples)
semantic_sim_map = {}
fp = open(SIMILARITY_MAP, 'w')
fp.write("uniprot_a\tuniprot_b\tcc_ss\tbp_ss\tmf_ss\n")
for p1, p2, cc_ss, bp_ss, mf_ss in sem_sim_tuples:
assert (p1, p2) not in semantic_sim_map
assert (p2, p1) not in semantic_sim_map
semantic_sim_map[(p1, p2)] = np.asarray([cc_ss, bp_ss, mf_ss])
fp.write('{}\t{}\t{}\t{}\t{}\n'.format(p1, p2, cc_ss, bp_ss, mf_ss))
fp.close()
else:
print("Loading semantic similarities...")
fp = open(SIMILARITY_MAP, 'r')
fp.readline()
semantic_sim_map = {}
)
testing_stats_i_j = evaluate_model(
y=y_hprd,
y_pred=y_pred_hprd,
y_pred_prob=y_proba_hprd,
label=l,
statistics=testing_stats_i_j,
verbose=0
)
validation_stats_i.merge(validation_stats_i_j)
testing_stats_i.merge(testing_stats_i_j)
return validation_stats_i, testing_stats_i
containers = parallel_map(do_iteration, zip(range(iterations), seeds), n_jobs=n_jobs)
valid_containers = [containers[i][0] for i in range(iterations)]
test_containers = [containers[i][1] for i in range(iterations)]
for container in valid_containers:
validation_stats.merge(container)
for container in test_containers:
testing_stats.merge(container)
# --------------------- FINAL RESULTS ---------------------------- #
direc = tempfile.mkdtemp(prefix='LLDA-{}-'.format(date), dir='results/')
pickle.dump((validation_stats, testing_stats, config), open(direc + '/LLDA-statistics.pkl', 'w'))
results = open(direc + '/LLDA-results.txt', 'w')
results.write("\nRun Settings: \n")
stop_words=['go', '', ' '], binary=binary, lowercase=True
)
vectorizer.fit(training_fold['terms'].values)
# Fit an evaluate the performance of the classifier.
x_train = vectorizer.transform(training_fold['terms'].values)
y_train = np.asarray(training_fold[label].values, dtype=int)
x_valid = vectorizer.transform(validation_fold['terms'].values)
y_valid = np.asarray(validation_fold[label].values, dtype=int)
estimator.fit(x_train, y_train)
for t in thresholds:
y_pred = [int(p[1] >= t) for p in estimator.predict_proba(x_valid)]
precision = precision_score(y_valid, y_pred, labels=[0, 1], pos_label=1)
recall = recall_score(y_valid, y_pred, labels=[0, 1], pos_label=1)
f1 = f1_score(y_valid, y_pred, labels=[0, 1], pos_label=1)
statistics_l.update_statistics(label=t, s_type='Precision', data=precision)
statistics_l.update_statistics(label=t, s_type='Recall', data=recall)
statistics_l.update_statistics(label=t, s_type='F1-Score', data=f1)
statistics_l.frame()['reaction'] = label
return statistics_l
containers = parallel_map(pr_curve, range(len(labels)), n_jobs=n_jobs)
dataframe = pd.concat([c.frame() for c in containers], ignore_index=True)
print(dataframe, len(dataframe))
dataframe.to_csv('analysis/pr_rc_curves.tsv', sep='\t', index=False)
def build_data_frame(ppi_file, obo_file, accession_to_feature_file, induce, fill_na, cache, n_jobs):
"""
Loads each tsv file containing a feature set (such as float similarity scores or accessions) into a pandas
dataframe and then attempts to create binary vector/bag of words representations of textual accesion
features. Finally combines each binary/numerical vector into a single feature vector along with it's label.
@param ppi_file: Directory to look for feature files.
@param obo_file: Path to obo file.
@param accession_to_feature_file: Path to accession-feature map stored in tsv.
@param induce: True to induce GO terms.
@param fill_na: Value to fill NA with. Best to use np.NaN.
@param cache: File to save dataframe to.
@return: DataFrame.
"""
print("Building dataframe...")
dag = ontology.load_go_dag(obo_file)
# Create the blank dataframe to which we will attach our data to.
labels = get_labels_from_file('data/labels.tsv')
od = Od({
'uniprot': [],
'uniprot_a': [],
'uniprot_b': [],
'go': [],
'go_cc': [],
'go_bp': [],
'go_mf': [],
'induced_go': [],
'induced_go_cc': [],
'induced_go_bp': [],
'induced_go_mf': [],
'ipr': [],
'pfam': [],
'sim': [],
'label': []
})
columns = od.keys() + labels
# This will be for quick accessing of data binary labels for BR methods.
# initialise these label to a null value.
for l in labels:
od[l] = -1
# Iterate through each ppi in the supplied file.
fp = open(ppi_file, 'r')
fp.readline() # assumes header exists for internal format
def do_line(line):
xs = line.strip().split('\t')
p1 = xs[0].strip()
p2 = xs[1].strip()
reaction_type = xs[2].strip()
reaction_types = [x.lower() for x in reaction_type.split(',')]
# Maybe use resnik or something.
cc_ss = float( xs[3].strip() )
bp_ss = float( xs[4].strip() )
mf_ss = float( xs[5].strip() )
terms = compute_features([p1, p2], induce, accession_to_feature_file, fill_na, dag)
od = Od({
'uniprot': [(p1, p2)],
'uniprot_a': [p1],
'uniprot_b': [p2],
'go': [terms['go']],
'go_cc': [terms['go_cc']],
'go_bp': [terms['go_bp']],
'go_mf': [terms['go_mf']],
'induced_go': [terms['induced_go']],
'induced_go_cc': [terms['induced_go_cc']],
'induced_go_bp': [terms['induced_go_bp']],
'induced_go_mf': [terms['induced_go_mf']],
'ipr': [terms['ipr']],
'pfam': [terms['pfam']],
'sim': [csr_matrix([cc_ss, bp_ss, mf_ss])],
'label': [reaction_type]
})
# Iterate and check which labels are present in reaction_type.
# Order of traversal is important here.
for l in labels:
if l.lower() in reaction_types:
od[l] = 1
else:
od[l] = 0
# Concatenate the dataframes
df_new = pd.DataFrame(
od, dtype='object',
columns=columns
)
return df_new
try:
df_rows = parallel_map(do_line, fp, n_jobs=n_jobs)
except KeyboardInterrupt:
sys.exit(0)
df = pd.concat(df_rows, ignore_index=True)
df = df.reset_index(); del df['index']
pickle.dump(df, open(cache, 'w'))
return df
df_valid=validation_fold,
df_test=test_df,
binary=binary,
connectivity='full',
vectorizer_method=vectorizer_method
)
validation_stats_i.merge(stats_valid)
testing_stats_i.merge(stats_test)
log.write('Iteration {}\n'.format(i))
validation_stats_i.write(log, 'a')
testing_stats_i.write(log, 'a')
return validation_stats_i, testing_stats_i
containers = parallel_map(do_iteration, range(iterations), n_jobs=n_jobs)
valid_containers = [containers[i][0] for i in range(iterations)]
test_containers = [containers[i][1] for i in range(iterations)]
for container in valid_containers:
validation_stats.merge(container)
for container in test_containers:
testing_stats.merge(container)
# --------------------- FINAL RESULTS ---------------------------- #
direc = tempfile.mkdtemp(prefix='{}-{}-'.format('CRF', date), dir='results/')
pickle.dump((validation_stats, testing_stats, config), open(direc + '/{}-statistics.pkl'.format('CRF'), 'w'))
results = open(direc + '/{}-results.txt'.format('CRF'), 'w')
results.write("\nRun Settings: \n")