from __future__ import print_function
import sys
from os.path import dirname, isfile, join as path_join
import numpy as np
from scipy.stats import pearsonr
from sklearn.metrics import r2_score
import warnings
try:
import compiledtrees
except ImportError:
compiledtrees = None
from oddt import random_seed
from oddt.metrics import rmse, standard_deviation_error
from oddt.scoring import scorer, ensemble_descriptor
from oddt.scoring.models.regressors import randomforest
from oddt.scoring.descriptors import (close_contacts_descriptor,
oddt_vina_descriptor)
# numpy after pickling gives Runtime Warnings
warnings.simplefilter("ignore", RuntimeWarning)
# RF-Score settings
ligand_atomic_nums = [6, 7, 8, 9, 15, 16, 17, 35, 53]
protein_atomic_nums = [6, 7, 8, 16]
cutoff = 12
[docs]class rfscore(scorer):
def __init__(self, protein=None, n_jobs=-1, version=1, spr=0, **kwargs):
"""Scoring function implementing RF-Score variants. It predicts the
binding affinity (pKi/d) of ligand in a complex utilizng simple
descriptors (close contacts of atoms <12A) with sophisticated
machine-learning model (random forest). The third variand supplements
those contacts with Vina partial scores. For futher details see RF-Score
publications v1[1]_, v2[2]_, v3[3]_.
Parameters
----------
protein : oddt.toolkit.Molecule object
Receptor for the scored ligands
n_jobs: int (default=-1)
Number of cores to use for scoring and training. By default (-1)
all cores are allocated.
version: int (default=1)
Scoring function variant. The deault is the simplest one (v1).
spr: int (default=0)
The minimum number of contacts in each pair of atom types in
the training set for the column to be included in training.
This is a way of removal of not frequent and empty contacts.
References
----------
.. [1] Ballester PJ, Mitchell JBO. A machine learning approach to
predicting protein-ligand binding affinity with applications to
molecular docking. Bioinformatics. 2010;26: 1169-1175.
doi:10.1093/bioinformatics/btq112
.. [2] Ballester PJ, Schreyer A, Blundell TL. Does a more precise
chemical description of protein-ligand complexes lead to more
accurate prediction of binding affinity? J Chem Inf Model. 2014;54:
944-955. doi:10.1021/ci500091r
.. [3] Li H, Leung K-S, Wong M-H, Ballester PJ. Improving AutoDock Vina
Using Random Forest: The Growing Accuracy of Binding Affinity
Prediction by the Effective Exploitation of Larger Data Sets.
Mol Inform. WILEY-VCH Verlag; 2015;34: 115-126.
doi:10.1002/minf.201400132
"""
self.protein = protein
self.n_jobs = n_jobs
self.version = version
self.spr = spr
if version == 1:
cutoff = 12
mtry = 6
descriptors = close_contacts_descriptor(
protein,
cutoff=cutoff,
protein_types=protein_atomic_nums,
ligand_types=ligand_atomic_nums)
elif version == 2:
cutoff = np.array([0, 2, 4, 6, 8, 10, 12])
mtry = 14
descriptors = close_contacts_descriptor(
protein,
cutoff=cutoff,
protein_types=protein_atomic_nums,
ligand_types=ligand_atomic_nums)
elif version == 3:
cutoff = 12
mtry = 6
cc = close_contacts_descriptor(
protein,
cutoff=cutoff,
protein_types=protein_atomic_nums,
ligand_types=ligand_atomic_nums)
vina_scores = ['vina_gauss1',
'vina_gauss2',
'vina_repulsion',
'vina_hydrophobic',
'vina_hydrogen',
'vina_num_rotors']
vina = oddt_vina_descriptor(protein, vina_scores=vina_scores)
descriptors = ensemble_descriptor((vina, cc))
model = randomforest(n_estimators=500,
oob_score=True,
n_jobs=n_jobs,
max_features=mtry,
bootstrap=True,
min_samples_split=6,
**kwargs)
super(rfscore, self).__init__(model, descriptors,
score_title='rfscore_v%i' % self.version)
[docs] def gen_training_data(self,
pdbbind_dir,
pdbbind_versions=(2007, 2012, 2013, 2014, 2015, 2016),
home_dir=None,
use_proteins=False):
if home_dir is None:
home_dir = dirname(__file__) + '/RFScore'
filename = path_join(home_dir, 'rfscore_descs_v%i.csv' % self.version)
super(rfscore, self)._gen_pdbbind_desc(
pdbbind_dir=pdbbind_dir,
pdbbind_versions=pdbbind_versions,
desc_path=filename,
use_proteins=use_proteins,
opt={'b': None},
)
[docs] def train(self, home_dir=None, sf_pickle=None, pdbbind_version=2016):
if not home_dir:
home_dir = dirname(__file__) + '/RFScore'
desc_path = path_join(home_dir, 'rfscore_descs_v%i.csv' % self.version)
super(rfscore, self)._load_pdbbind_desc(desc_path,
pdbbind_version=pdbbind_version)
# remove sparse dimentions
if self.spr > 0:
self.mask = (self.train_descs > self.spr).any(axis=0)
if self.mask.sum() > 0:
self.train_descs = self.train_descs[:, self.mask]
self.test_descs = self.test_descs[:, self.mask]
# make nets reproducible
random_seed(1)
self.model.fit(self.train_descs, self.train_target)
print('Training RFScore v%i on PDBBind v%i'
% (self.version, pdbbind_version), file=sys.stderr)
sets = [
('Test', self.model.predict(self.test_descs), self.test_target),
('Train', self.model.predict(self.train_descs), self.train_target),
('OOB', self.model.oob_prediction_, self.train_target)]
for name, pred, target in sets:
if len(target) < 3:
print('There are less than 3 values to predict, skipping.', file=sys.stderr)
continue
print('%s set:' % name,
'R2_score: %.4f' % r2_score(target, pred),
'Rp: %.4f' % pearsonr(target, pred)[0],
'RMSE: %.4f' % rmse(target, pred),
'SD: %.4f' % standard_deviation_error(target, pred),
sep='\t', file=sys.stderr)
# compile trees
if compiledtrees is not None:
try:
print('Compiling Random Forest using sklearn-compiledtrees',
file=sys.stderr)
self.model = compiledtrees.CompiledRegressionPredictor(
self.model, n_jobs=self.n_jobs)
except Exception as e:
print('Failed to compile Random Forest with exception: %s' % e,
file=sys.stderr)
print('Continuing without compiled RF.', file=sys.stderr)
if sf_pickle is None:
return self.save('RFScore_v%i_pdbbind%i.pickle'
% (self.version, pdbbind_version))
else:
return self.save(sf_pickle)
[docs] @classmethod
def load(self, filename=None, version=1, pdbbind_version=2016):
if filename is None:
fname = 'RFScore_v%i_pdbbind%i.pickle' % (version, pdbbind_version)
for f in [fname, path_join(dirname(__file__), fname)]:
if isfile(f):
filename = f
break
else:
print('No pickle, training new scoring function.',
file=sys.stderr)
rf = rfscore(version=version)
filename = rf.train(sf_pickle=filename,
pdbbind_version=pdbbind_version)
return scorer.load(filename)