Initial commit of pose scorer and featurized dataset

deepchem/dock/__init__.py

@@ -6,3 +6,4 @@
 from __future__ import unicode_literals
 
 from deepchem.dock.pose_generation import VinaPoseGenerator
+from deepchem.dock.pose_scoring import PoseScorer

deepchem/dock/pose_scoring.py

@@ -0,0 +1,42 @@
+"""
+Scores protein-ligand poses using DeepChem.
+"""
+from __future__ import print_function
+from __future__ import division
+from __future__ import unicode_literals
+
+__author__ = "Bharath Ramsundar"
+__copyright__ = "Copyright 2016, Stanford University"
+__license__ = "GPL"
+
+import numpy as np
+import os
+import tempfile
+from deepchem.feat import GridFeaturizer
+from deepchem.data import NumpyDataset
+from subprocess import call
+
+class PoseScorer(object):
+
+  def __init__(self, model, feat="grid"):
+    """Initializes a pose-scorer."""
+    self.model = model
+    if feat == "grid":
+      self.featurizer = GridFeaturizer(
+          voxel_width=16.0, feature_types="voxel_combined",
+          # TODO(rbharath, enf): Figure out why pi_stack is slow and cation_pi
+          # causes segfaults.
+          #voxel_feature_types=["ecfp", "splif", "hbond", "pi_stack", "cation_pi",
+          #"salt_bridge"], ecfp_power=9, splif_power=9,
+          voxel_feature_types=["ecfp", "splif", "hbond", "salt_bridge"],
+          ecfp_power=9, splif_power=9,
+          parallel=True, flatten=True)
+    else:
+      raise ValueError("feat not defined.")
+
+  def score(self, protein_file, ligand_file):
+    """Returns a score for a protein/ligand pair."""
+    features = self.featurizer.featurize_complexes([ligand_file], [protein_file])
+    dataset = NumpyDataset(X=features, y=None, w=None, ids=None)
+    score = self.model.predict(dataset)
+    return score

deepchem/dock/tests/test_pose_scoring.py

@@ -0,0 +1,37 @@
+"""
+Tests for Pose Scoring 
+"""
+from __future__ import print_function
+from __future__ import division
+from __future__ import unicode_literals
+
+__author__ = "Bharath Ramsundar"
+__copyright__ = "Copyright 2016, Stanford University"
+__license__ = "GPL"
+
+import unittest
+import tempfile
+import os
+import shutil
+import numpy as np
+import deepchem as dc
+from sklearn.ensemble import RandomForestRegressor
+from subprocess import call
+
+class TestPoseScoring(unittest.TestCase):
+  """
+  Does sanity checks on pose generation. 
+  """
+
+  def test_pose_scorer_init(self):
+    """Tests that pose-score works."""
+    call("wget http://deepchem.io.s3-website-us-west-1.amazonaws.com/featurized_datasets/core_grid.tar.gz".split())
+    call("tar -zxvf core_grid.tar.gz".split())
+    core_dataset = dc.data.DiskDataset("core_grid/")
+
+    sklearn_model = RandomForestRegressor(n_estimators=10)
+    model = dc.models.SklearnModel(sklearn_model)
+    print("About to fit model on core set")
+    model.fit(core_dataset)
+
+    pose_scorer = dc.dock.PoseScorer(model, feat="grid")

examples/pdbbind/get_featurized_pdbbind.sh

@@ -0,0 +1,8 @@
+echo "Pulling featurized core pdbbind dataset from deepchem"
+wget http://deepchem.io.s3-website-us-west-1.amazonaws.com/featurized_datasets/core_grid.tar.gz
+echo "Extracting core pdbbind"
+tar -zxvf core_grid.tar.gz
+echo "Pulling featurized refined pdbbind dataset from deepchem"
+wget http://deepchem.io.s3-website-us-west-1.amazonaws.com/featurized_datasets/refined_grid.tar.gz
+echo "Extracting refined pdbbind"
+tar -zxvf refined_grid.tar.gz

examples/pdbbind/pdbbind_datasets.py

@@ -11,18 +11,29 @@
 import pandas as pd
 import shutil
 import time
+import re
 from rdkit import Chem
 import deepchem as dc
 
 def load_pdbbind_labels(labels_file):
   """Loads pdbbind labels as dataframe"""
+  # Some complexes have labels but no PDB files. Filter these manually
+  missing_pdbs = ["1d2v", "1jou", "1s8j", "1cam", "4mlt", "4o7d"]
   contents = []
   with open(labels_file) as f:
     for line in f:
       if line.startswith("#"):
         continue
       else:
-        contents.append(line.split())
+        # Some of the ligand-names are of form (FMN ox). Use regex
+        # to merge into form (FMN-ox)
+        p = re.compile('\(([^\)\s]*) ([^\)\s]*)\)')
+        line = p.sub('(\\1-\\2)', line)
+        elts = line.split()
+        # Filter if missing PDB files
+        if elts[0] in missing_pdbs:
+          continue
+        contents.append(elts)
   contents_df = pd.DataFrame(
       contents,
       columns=("PDB code", "resolution", "release year", "-logKd/Ki", "Kd/Ki",
@@ -86,10 +97,15 @@ def featurize_pdbbind(data_dir=None, feat="grid", subset="core"):
   features = []
   feature_len = None
   y_inds = []
+  missing_pdbs = []
   time1 = time.time()
   for ind, pdb_code in enumerate(ids):
     print("Processing complex %d, %s" % (ind, str(pdb_code)))
     pdb_subdir = os.path.join(pdbbind_dir, pdb_code)
+    if not os.path.exists(pdb_subdir):
+      print("%s is missing!" % pdb_subdir)
+      missing_pdbs.append(pdb_subdir)
+      continue
     computed_feature = compute_pdbbind_features(
         featurizer, pdb_subdir, pdb_code)
     if feature_len is None:
@@ -101,6 +117,8 @@ def featurize_pdbbind(data_dir=None, feat="grid", subset="core"):
     features.append(computed_feature)
   time2 = time.time()
   print("TIMING: PDBBind Featurization took %0.3f s" % (time2-time1))
+  print("missing_pdbs")
+  print(missing_pdbs)
   y = y[y_inds]
   X = np.vstack(features)
   w = np.ones_like(y)
@@ -114,8 +132,7 @@ def load_pdbbind_grid(split="index", feat="grid", subset="core"):
   transformers = []
 
   splitters = {'index': dc.splits.IndexSplitter(),
-               'random': dc.splits.RandomSplitter(),
-               'scaffold': dc.splits.ScaffoldSplitter()}
+               'random': dc.splits.RandomSplitter()}
   splitter = splitters[split]
   train, valid, test = splitter.train_valid_test_split(dataset)