Merge pull request #57 from evanfeinberg/master (ef26e227) · Commits · 钟慕尧 / deepchem

datasets/construct_pdbbind_df.py

0 → 100644

+93 −0

Original line number	Diff line number	Diff line
		"""
		Contains methods for generating a pdbbind dataset mapping
		complexes (protein + ligand) to experimental binding measurement.
		"""
		import pickle
		import os
		import pandas as pd
		from rdkit import Chem
		from glob import glob
		import re


		def extract_labels(pdbbind_label_file):
		"""Extract labels from pdbbind label file."""
		assert os.path.isfile(pdbbind_label_file)
		labels = {}
		with open(pdbbind_label_file) as f:
		content = f.readlines()
		for line in content:
		if line[0] == "#":
		continue
		line = line.split()
		# lines in the label file have format
		# PDB-code Resolution Release-Year -logKd Kd reference ligand-name
		#print line[0], line[3]
		labels[line[0]] = line[3]
		return labels

		def construct_df(pdb_stem_directory, pdbbind_label_file, pdbbind_df_pkl):
		"""
		Takes as input a stem directory containing subdirectories with ligand
		and protein pdb/mol2 files, a pdbbind_label_file containing binding
		assay data for the co-crystallized ligand in each pdb file,
		and a pdbbind_df_pkl to which will be saved a pandas DataFrame
		where each row contains a pdb_id, smiles string, unique complex id,
		ligand pdb as a list of strings per line in file, protein pdb as a list
		of strings per line in file, ligand mol2 as a list of strings per line in
		mol2 file, and a "label" containing the experimental measurement.
		"""
		labels = extract_labels(pdbbind_label_file)
		df_rows = []
		os.chdir(pdb_stem_directory)
		pdb_directories = [pdb.replace('/', '') for pdb in glob('*/')]

		for pdb_dir in pdb_directories:
		print "About to extract ligand and protein input files"
		pdb_id = os.path.basename(pdb_dir)
		ligand_pdb = None
		protein_pdb = None
		for f in os.listdir(pdb_dir):
		if re.search("_ligand_hyd.pdb$", f):
		ligand_pdb = f
		elif re.search("_protein_hyd.pdb$", f):
		protein_pdb = f
		elif re.search("_ligand.mol2$", f):
		ligand_mol2 = f

		print "Extracted Input Files:"
		print (ligand_pdb, protein_pdb, ligand_mol2)
		if not ligand_pdb or not protein_pdb or not ligand_mol2:
		raise ValueError("Required files not present for %s" % pdb_dir)
		ligand_pdb_path = os.path.join(pdb_dir, ligand_pdb)
		protein_pdb_path = os.path.join(pdb_dir, protein_pdb)
		ligand_mol2_path = os.path.join(pdb_dir, ligand_mol2)

		with open(protein_pdb_path, "rb") as f:
		protein_pdb_lines = f.readlines()

		with open(ligand_pdb_path, "rb") as f:
		ligand_pdb_lines = f.readlines()

		try:
		with open(ligand_mol2_path, "rb") as f:
		ligand_mol2_lines = f.readlines()
		except:
		ligand_mol2_lines = []

		print "About to compute ligand smiles string."
		ligand_mol = Chem.MolFromPDBFile(ligand_pdb_path)
		if ligand_mol is None:
		continue
		smiles = Chem.MolToSmiles(ligand_mol)
		complex_id = "%s%s" % (pdb_id, smiles)
		label = labels[pdb_id]
		df_rows.append([pdb_id, smiles, complex_id, protein_pdb_lines,
		ligand_pdb_lines, ligand_mol2_lines, label])

		pdbbind_df = pd.DataFrame(df_rows, columns=('pdb_id', 'smiles', 'complex_id',
		'protein_pdb', 'ligand_pdb',
		'ligand_mol2', 'label'))

		with open(pdbbind_df_pkl, "wb") as f:
		pickle.dump(pdbbind_df, f)

datasets/pdbbind_core_df.pkl.gz

0 → 100644

+32.1 MiB

File added.

No diff preview for this file type.

View file

deepchem/models/init.py

+5 −2

Original line number	Diff line number	Diff line
		@@ -128,11 +128,12 @@ class Model(object):
		for epoch in range(self.model_params["nb_epoch"]):
		print("Starting epoch %s" % str(epoch+1))
		for i, (X, y, w, _) in enumerate(dataset.itershards()):
		print("Training on batch-%s/epoch-%s" % (str(i+1), str(epoch+1)))
		print("Training on shard-%s/epoch-%s" % (str(i+1), str(epoch+1)))
		nb_sample = np.shape(X)[0]
		interval_points = np.linspace(
		0, nb_sample, np.ceil(float(nb_sample)/batch_size)+1, dtype=int)
		for j in range(len(interval_points)-1):
		print("Training on batch-%s/shard-%s/epoch-%s" % (str(j+1), str(i+1), str(epoch+1)))
		indices = range(interval_points[j], interval_points[j+1])
		X_batch = X[indices, :]
		y_batch = y[indices]
		@@ -160,7 +161,9 @@ class Model(object):
		y_preds = []
		for j in range(len(interval_points)-1):
		indices = range(interval_points[j], interval_points[j+1])
		y_preds.append(self.predict_on_batch(X[indices, :]))
		y_pred_on_batch = self.predict_on_batch(X[indices, :]).reshape((len(indices),len(task_names)))
		y_preds.append(y_pred_on_batch)

		y_pred = np.concatenate(y_preds)
		y_pred = np.reshape(y_pred, np.shape(y))

deepchem/scripts/modeler.py

+30 −17

Original line number	Diff line number	Diff line
		@@ -45,8 +45,14 @@ def add_featurize_group(featurize_cmd):
		"--threshold", type=float, default=None,
		help="If specified, will be used to binarize real-valued target-fields.")
		featurize_group.add_argument(
		"--parallel", type=float, default=None,
		help="Use multiprocessing will be used to parallelize featurization.")
		"--protein-pdb-field", type=str, default=None,
		help="Name of field holding protein pdb.")
		featurize_group.add_argument(
		"--ligand-pdb-field", type=str, default=None,
		help="Name of field holding ligand pdb.")
		featurize_group.add_argument(
		"--ligand-mol2-field", type=str, default=None,
		help="Name of field holding ligand mol2.")

		def add_transforms_group(cmd):
		"""Adds flags for data transforms."""
		@@ -62,7 +68,7 @@ def add_transforms_group(cmd):
		"to mean no transforms are required.")
		transform_group.add_argument(
		"--feature-types", nargs="+", required=1,
		choices=["user-specified-features", "ECFP", "RDKIT-descriptors"],
		choices=["user-specified-features", "ECFP", "RDKIT-descriptors", "NNScore"],
		help="Featurizations of data to use.\n"
		"'features' denotes user-defined features.\n"
		"'fingerprints' denotes ECFP fingeprints.\n"
		@@ -205,7 +211,8 @@ def create_model(args):
		featurize_inputs(
		feature_dir, data_dir, args.input_files, args.user_specified_features,
		args.tasks, args.smiles_field, args.split_field, args.id_field,
		args.threshold, args.parallel)
		args.threshold, args.protein_pdb_field,
		args.ligand_pdb_field, args.ligand_mol2_field)

		if args.generate_dataset:
		print("+++++++++++++++++++++++++++++++++")
		@@ -273,9 +280,14 @@ def parse_args(input_args=None):
		add_model_command(subparsers)
		return parser.parse_args(input_args)

		def shard_inputs(input_file):
		input_file_no_ext = os.path.splitext(input_file)


		def featurize_inputs(feature_dir, data_dir, input_files,
		user_specified_features, tasks, smiles_field,
		split_field, id_field, threshold, parallel):
		split_field, id_field, threshold, protein_pdb_field,
		ligand_pdb_field, ligand_mol2_field):

		"""Allows for parallel data featurization."""
		featurize_input_partial = partial(featurize_input,
		@@ -285,13 +297,11 @@ def featurize_inputs(feature_dir, data_dir, input_files,
		smiles_field=smiles_field,
		split_field=split_field,
		id_field=id_field,
		threshold=threshold)
		threshold=threshold,
		protein_pdb_field=protein_pdb_field,
		ligand_pdb_field=ligand_pdb_field,
		ligand_mol2_field=ligand_mol2_field)

		if parallel:
		pool = mp.Pool(int(mp.cpu_count()/2))
		pool.map(featurize_input_partial, input_files)
		pool.terminate()
		else:
		for input_file in input_files:
		featurize_input_partial(input_file)

		@@ -302,18 +312,21 @@ def featurize_inputs(feature_dir, data_dir, input_files,
		FeaturizedSamples(samples_dir, dataset_files)

		def featurize_input(input_file, feature_dir, user_specified_features, tasks,
		smiles_field, split_field, id_field, threshold):
		smiles_field, split_field, id_field, threshold, protein_pdb_field,
		ligand_pdb_field, ligand_mol2_field):
		"""Featurizes raw input data."""
		featurizer = DataFeaturizer(tasks=tasks,
		smiles_field=smiles_field,
		split_field=split_field,
		id_field=id_field,
		threshold=threshold,
		protein_pdb_field=protein_pdb_field,
		ligand_pdb_field=ligand_pdb_field,
		ligand_mol2_field=ligand_mol2_field,
		user_specified_features=user_specified_features,
		verbose=True)
		out = os.path.join(
		feature_dir, "%s.joblib" %(os.path.splitext(os.path.basename(input_file))[0]))
		featurizer.featurize(input_file, FeaturizedSamples.feature_types, out)

		featurizer.featurize(input_file, FeaturizedSamples.feature_types, feature_dir)

		def train_test_split(input_transforms, output_transforms,
		feature_types, splittype, data_dir):

deepchem/utils/dataset.py

+2 −1

Original line number	Diff line number	Diff line
		@@ -314,7 +314,8 @@ def _df_to_numpy(df, feature_types):
		sorted_ids = df["mol_id"]

		# Set missing data to have weight zero
		missing = (y == "")
		missing = (y.astype(object) == "")

		y[missing] = 0.
		w[missing] = 0.

Admin message