Molecular docking, scoring, and virtual screening play an increasingly important role in computer-aided drug discovery. Scoring functions (SFs) are typically employed to predict the binding conformation (docking task), binding affinity (scoring task), and binary activity level (screening task) of ligands against a critical protein target in a disease's pathway. In most molecular docking software packages available today, a generic binding affinity-based (BA-based) SF is invoked for all three tasks to solve three different, but related, prediction problems. The limited predictive accuracies of such SFs in these three tasks has been a major roadblock toward cost-effective drug discovery. Therefore, in this work, we develop BT-Score, an ensemble machine-learning (ML) SF of boosted decision trees and thousands of predictive descriptors to estimate BA. BT-Score reproduced BA of out-of-sample test complexes with correlation of 0.825. Even with this high accuracy in the scoring task, we demonstrate that the docking and screening performance of BT-Score and other BA-based SFs is far from ideal. This has motivated us to build two task-specific ML SFs for the docking and screening problems. We propose BT-Dock, a boosted-tree ensemble model trained on a large number of native and computer-generated ligand conformations and optimized to predict binding poses explicitly. This model has shown an average improvement of 25% over its BA-based counterparts in different ligand pose prediction scenarios. Similar improvement has also been obtained by our screening-based SF, BT-Screen, which directly models the ligand activity labeling task as a classification problem. BT-Screen is trained on thousands of active and inactive protein-ligand complexes to optimize it for finding real actives from databases of ligands not seen in its training set. In addition to the three task-specific SFs, we propose a novel multi-task deep neural network (MT-Net) that is trained on data from the three tasks to simultaneously predict binding poses, affinities, and activity levels. We show that the performance of MT-Net is superior to conventional SFs and on a par with or better than models based on single-task neural networks.

Accurately predicting the binding affinities of large diverse sets of protein-ligand complexes efficiently is a key challenge in computational biomolecular science, with applications in drug discovery, chemical biology, and structural biology. Since a scoring function (SF) is used to score, rank, and identify potential drug leads, the fidelity with which it predicts the affinity of a ligand candidate for a protein's binding site has a significant bearing on the accuracy of virtual screening. Despite intense efforts in developing conventional SFs, which are either force-field based, knowledge-based, or empirical, their limited predictive accuracy has been a major roadblock toward cost-effective drug discovery. Therefore, in this work, we explore a range of novel SFs employing different machine-learning (ML) approaches in conjunction with a variety of physicochemical and geometrical features characterizing protein-ligand complexes. We assess the scoring accuracies of these new ML SFs as well as those of conventional SFs in the context of the 2007 and 2010 PDBbind benchmark datasets on both diverse and protein-family-specific test sets. We also investigate the influence of the size of the training dataset and the type and number of features used on scoring accuracy. We find that the best performing ML SF has a Pearson correlation coefficient of 0.806 between predicted and measured binding affinities compared to 0.644 achieved by a state-of-the-art conventional SF. We also find that ML SFs benefit more than their conventional counterparts from increases in the number of features and the size of training dataset. In addition, they perform better on novel proteins that they were never trained on before.

BackgroundAccurately predicting the binding affinities of large sets of protein-ligand complexes is a key challenge in computational biomolecular science, with applications in drug discovery, chemical biology, and structural biology. Since a scoring function (SF) is used to score, rank, and identify drug leads, the fidelity with which it predicts the affinity of a ligand candidate for a protein's binding site has a significant bearing on the accuracy of virtual screening. Despite intense efforts in developing conventional SFs, which are either force-field based, knowledge-based, or empirical, their limited predictive power has been a major roadblock toward cost-effective drug discovery. Therefore, in this work, we present novel SFs employing a large ensemble of neural networks (NN) in conjunction with a diverse set of physicochemical and geometrical features characterizing protein-ligand complexes to predict binding affinity.ResultsWe assess the scoring accuracies of two new ensemble NN SFs based on bagging (BgN-Score) and boosting (BsN-Score), as well as those of conventional SFs in the context of the 2007 PDBbind benchmark that encompasses a diverse set of high-quality protein families. We find that BgN-Score and BsN-Score have more than 25% better Pearson's correlation coefficient (0.804 and 0.816 vs. 0.644) between predicted and measured binding affinities compared to that achieved by a state-of-the-art conventional SF. In addition, these ensemble NN SFs are also at least 19% more accurate (0.804 and 0.816 vs. 0.675) than SFs based on a single neural network that has been traditionally used in drug discovery applications. We further find that ensemble models based on NNs surpass SFs based on the decision-tree ensemble technique Random Forests.ConclusionsEnsemble neural networks SFs, BgN-Score and BsN-Score, are the most accurate in predicting binding affinity of protein-ligand complexes among the considered SFs. Moreover, their accuracies are even higher when they are used to predict binding affinities of protein-ligand complexes that are related to their training sets.

BackgroundMolecular docking is a widely-employed method in structure-based drug design. An essential component of molecular docking programs is a scoring function (SF) that can be used to identify the most stable binding pose of a ligand, when bound to a receptor protein, from among a large set of candidate poses. Despite intense efforts in developing conventional SFs, which are either force-field based, knowledge-based, or empirical, their limited docking power (or ability to successfully identify the correct pose) has been a major impediment to cost-effective drug discovery. Therefore, in this work, we explore a range of novel SFs employing different machine-learning (ML) approaches in conjunction with physicochemical and geometrical features characterizing protein-ligand complexes to predict the native or near-native pose of a ligand docked to a receptor protein's binding site. We assess the docking accuracies of these new ML SFs as well as those of conventional SFs in the context of the 2007 PDBbind benchmark dataset on both diverse and homogeneous (protein-family-specific) test sets. Further, we perform a systematic analysis of the performance of the proposed SFs in identifying native poses of ligands that are docked to novel protein targets.Results and conclusionWe find that the best performing ML SF has a success rate of 80% in identifying poses that are within 1 Å root-mean-square deviation from the native poses of 65 different protein families. This is in comparison to a success rate of only 70% achieved by the best conventional SF, ASP, employed in the commercial docking software GOLD. In addition, the proposed ML SFs perform better on novel proteins that they were never trained on before. We also observed steady gains in the performance of these scoring functions as the training set size and number of features were increased by considering more protein-ligand complexes and/or more computationally-generated poses for each complex.

No abstract