Evaluation of ligand-binding affinity using polynomial empirical scoring functions

Computational Prediction of Binding Affinity for CDK2-ligand Complexes. A Protein Target for Cancer Drug Discovery

BACKGROUND

CDK2 participates in the control of eukaryotic cell-cycle progression. Due to the great interest in CDK2 for drug development and the relative easiness in crystallizing this enzyme, we have over 400 structural studies focused on this protein target. This structural data is the basis for the development of computational models to estimate CDK2-ligand binding affinity.

OBJECTIVE

This work focuses on the recent developments in the application of supervised machine learning modeling to develop scoring functions to predict the binding affinity of CDK2.

METHOD

We employed the structures available at the protein data bank and the ligand information accessed from the BindingDB, Binding MOAD, and PDBbind to evaluate the predictive performance of machine learning techniques combined with physical modeling used to calculate binding affinity. We compared this hybrid methodology with classical scoring functions available in docking programs.

RESULTS

Our comparative analysis of previously published models indicated that a model created using a combination of a mass-spring system and cross-validated Elastic Net to predict the binding affinity of CDK2-inhibitor complexes outperformed classical scoring functions available in AutoDock4 and AutoDock Vina.

CONCLUSION

All studies reviewed here suggest that targeted machine learning models are superior to classical scoring functions to calculate binding affinities. Specifically for CDK2, we see that the combination of physical modeling with supervised machine learning techniques exhibits improved predictive performance to calculate the protein-ligand binding affinity. These results find theoretical support in the application of the concept of scoring function space.

Machine Learning-Based Scoring Functions, Development and Applications with SAnDReS

BACKGROUND

Analysis of atomic coordinates of protein-ligand complexes can provide three-dimensional data to generate computational models to evaluate binding affinity and thermodynamic state functions. Application of machine learning techniques can create models to assess protein-ligand potential energy and binding affinity. These methods show superior predictive performance when compared with classical scoring functions available in docking programs.

OBJECTIVE

Our purpose here is to review the development and application of the program SAnDReS. We describe the creation of machine learning models to assess the binding affinity of protein-ligand complexes.

METHODS

SAnDReS implements machine learning methods available in the scikit-learn library. This program is available for download at https://github.com/azevedolab/sandres. SAnDReS uses crystallographic structures, binding and thermodynamic data to create targeted scoring functions.

RESULTS

Recent applications of the program SAnDReS to drug targets such as Coagulation factor Xa, cyclin-dependent kinases and HIV-1 protease were able to create targeted scoring functions to predict inhibition of these proteins. These targeted models outperform classical scoring functions.

CONCLUSION

Here, we reviewed the development of machine learning scoring functions to predict binding affinity through the application of the program SAnDReS. Our studies show the superior predictive performance of the SAnDReS-developed models when compared with classical scoring functions available in the programs such as AutoDock4, Molegro Virtual Docker and AutoDock Vina.

Structural Basis for Inhibition of Enoyl-[Acyl Carrier Protein] Reductase (InhA) from Mycobacterium tuberculosis

Background::

The enzyme trans-enoyl-[acyl carrier protein] reductase (InhA) is a central protein for the development of antitubercular drugs. This enzyme is the target for the pro-drug isoniazid, which is catalyzed by the enzyme catalase-peroxidase (KatG) to become active.

Objective::

Our goal here is to review the studies on InhA, starting with general aspects and focusing on the recent structural studies, with emphasis on the crystallographic structures of complexes involving InhA and inhibitors.

Method::

We start with a literature review, and then we describe recent studies on InhA crystallographic structures. We use this structural information to depict protein-ligand interactions. We also analyze the structural basis for inhibition of InhA. Furthermore, we describe the application of computational methods to predict binding affinity based on the crystallographic position of the ligands.

Results::

Analysis of the structures in complex with inhibitors revealed the critical residues responsible for the specificity against InhA. Most of the intermolecular interactions involve the hydrophobic residues with two exceptions, the residues Ser 94 and Tyr 158. Examination of the interactions has shown that many of the key residues for inhibitor binding were found in mutations of the InhA gene in the isoniazid-resistant Mycobacterium tuberculosis. Computational prediction of the binding affinity for InhA has indicated a moderate uphill relationship with experimental values.

Conclusion::

Analysis of the structures involving InhA inhibitors shows that small modifications on these molecules could modulate their inhibition, which may be used to design novel antitubercular drugs specific for multidrug-resistant strains.

Docking with GemDock

GEMDOCK is a protein-ligand docking software that makes use of an elegant biologically inspired computational methodology based on the differential evolution algorithm. As any docking program, GEMDOCK has two major features to predict the binding of a small-molecule ligand to the binding site of a protein target: the search algorithm and the scoring function to evaluate the generated poses. The GEMDOCK scoring function uses a piecewise potential energy function integrated into the differential evolutionary algorithm. GEMDOCK has been applied to a wide range of protein systems with docking accuracy similar to other docking programs such as Molegro Virtual Docker, AutoDock4, and AutoDock Vina. In this chapter, we explain how to carry out protein-ligand docking simulations with GEMDOCK. We focus this tutorial on the protein target cyclin-dependent kinase 2.

SAnDReS: A Computational Tool for Docking

Since the early 1980s, we have witnessed considerable progress in the development and application of docking programs to assess protein-ligand interactions. Most of these applications had as a goal the identification of potential new binders to protein targets. Another remarkable progress is taking place in the determination of the structures of protein-ligand complexes, mostly using X-ray diffraction crystallography. Considering these developments, we have a favorable scenario for the creation of a computational tool that integrates into one workflow all steps involved in molecular docking simulations. We had these goals in mind when we developed the program SAnDReS. This program allows the integration of all computational features related to modern docking studies into one workflow. SAnDReS not only carries out docking simulations but also evaluates several docking protocols allowing the selection of the best approach for a given protein system. SAnDReS is a free and open-source (GNU General Public License) computational environment for running docking simulations. Here, we describe the combination of SAnDReS and AutoDock4 for protein-ligand docking simulations. AutoDock4 is a free program that has been applied to over a thousand receptor-ligand docking simulations. The dataset described in this chapter is available for downloading at https://github.com/azevedolab/sandres.

Van der Waals Potential in Protein Complexes

Van der Waals forces are determinants of the formation of protein-ligand complexes. Physical models based on the Lennard-Jones potential can estimate van der Waals interactions with considerable accuracy and with a computational complexity that allows its application to molecular docking simulations and virtual screening of large databases of small organic molecules. Several empirical scoring functions used to evaluate protein-ligand interactions approximate van der Waals interactions with the Lennard-Jones potential. In this chapter, we present the main concepts necessary to understand van der Waals interactions relevant to molecular recognition of a ligand by the binding pocket of a protein target. We describe the Lennard-Jones potential and its application to calculate potential energy for an ensemble of structures to highlight the main features related to the importance of this interaction for binding affinity.

Homology Modeling of Protein Targets with MODELLER

Homology modeling is a computational approach to generate three-dimensional structures of protein targets when experimental data about similar proteins are available. Although experimental methods such as X-ray crystallography and nuclear magnetic resonance spectroscopy successfully solved the structures of nearly 150,000 macromolecules, there is still a gap in our structural knowledge. We can fulfill this gap with computational methodologies. Our goal in this chapter is to explain how to perform homology modeling of protein targets for drug development. We choose as a homology modeling tool the program MODELLER. To illustrate its use, we describe how to model the structure of human cyclin-dependent kinase 3 using MODELLER. We explain the modeling procedure of CDK3 apoenzyme and the structure of this enzyme in complex with roscovitine.

Exploring the Scoring Function Space

In the analysis of protein-ligand interactions, two abstractions have been widely employed to build a systematic approach to analyze these complexes: protein and chemical spaces. The pioneering idea of the protein space dates back to 1970, and the chemical space is newer, later 1990s. With the progress of computational methodologies to create machine-learning models to predict the ligand-binding affinity, clearly there is a need for novel approaches to the problem of protein-ligand interactions. New abstractions are required to guide the conceptual analysis of the molecular recognition problem. Using a systems approach, we proposed to address protein-ligand scoring functions using the modern idea of the scoring function space. In this chapter, we describe the fundamental concept behind the scoring function space and how it has been applied to develop the new generation of targeted-scoring functions.

Machine Learning to Predict Binding Affinity

Recent progress in the development of scientific libraries with machine-learning techniques paved the way for the implementation of integrated computational tools to predict ligand-binding affinity. The prediction of binding affinity uses the atomic coordinates of protein-ligand complexes. These new computational tools made application of a broad spectrum of machine-learning techniques to study protein-ligand interactions possible. The essential aspect of these machine-learning approaches is to train a new computational model by using technologies such as supervised machine-learning techniques, convolutional neural network, and random forest to mention the most commonly applied methods. In this chapter, we focus on supervised machine-learning techniques and their applications in the development of protein-targeted scoring functions for the prediction of binding affinity. We discuss the development of the program SAnDReS and its application to the creation of machine-learning models to predict inhibition of cyclin-dependent kinase and HIV-1 protease. Moreover, we describe the scoring function space, and how to use it to explain the development of targeted scoring functions.

Hydrogen Bonds in Protein-Ligand Complexes

Fast and reliable evaluation of the hydrogen bond potential energy has a significant impact in the drug design and development since it allows the assessment of large databases of organic molecules in virtual screening projects focused on a protein of interest. Semi-empirical force fields implemented in molecular docking programs make it possible the evaluation of protein-ligand binding affinity where the hydrogen bond potential is a common term used in the calculation. In this chapter, we describe the concepts behind the programs used to predict hydrogen bond potential energy employing semi-empirical force fields as the ones available in the programs AMBER, AutoDock4, TreeDock, and ReplicOpter. We described here the 12-10 potential and applied it to evaluate the binding affinity for an ensemble of crystallographic structures for which experimental data about binding affinity are available.

Improving the accuracy of high-throughput protein-protein affinity prediction may require better training data

Background

One goal of structural biology is to understand how a protein’s 3-dimensional conformation determines its capacity to interact with potential ligands. In the case of small chemical ligands, deconstructing a static protein-ligand complex into its constituent atom-atom interactions is typically sufficient to rapidly predict ligand affinity with high accuracy (>70% correlation between predicted and experimentally-determined affinity), a fact that is exploited to support structure-based drug design. We recently found that protein-DNA/RNA affinity can also be predicted with high accuracy using extensions of existing techniques, but protein-protein affinity could not be predicted with >60% correlation, even when the protein-protein complex was available.

Methods

X-ray and NMR structures of protein-protein complexes, their associated binding affinities and experimental conditions were obtained from different binding affinity and structural databases. Statistical models were implemented using a generalized linear model framework, including the experimental conditions as new model features. We evaluated the potential for new features to improve affinity prediction models by calculating the Pearson correlation between predicted and experimental binding affinities on the training and test data after model fitting and after cross-validation. Differences in accuracy were assessed using two-sample t test and nonparametric Mann–Whitney U test.

Results

Here we evaluate a range of potential factors that may interfere with accurate protein-protein affinity prediction. We find that X-ray crystal resolution has the strongest single effect on protein-protein affinity prediction. Limiting our analyses to only high-resolution complexes (≤2.5 Å) increased the correlation between predicted and experimental affinity from 54 to 68% (p = 4.32x10⁻³). In addition, incorporating information on the experimental conditions under which affinities were measured (pH, temperature and binding assay) had significant effects on prediction accuracy. We also highlight a number of potential errors in large structure-affinity databases, which could affect both model training and accuracy assessment.

Conclusions

The results suggest that the accuracy of statistical models for protein-protein affinity prediction may be limited by the information present in databases used to train new models. Improving our capacity to integrate large-scale structural and functional information may be required to substantively advance our understanding of the general principles by which a protein’s structure determines its function.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-017-1533-z) contains supplementary material, which is available to authorized users.

Different combinations of atomic interactions predict protein-small molecule and protein-DNA/RNA affinities with similar accuracy

Interactions between proteins and other molecules play essential roles in all biological processes. Although it is widely held that a protein's ligand specificity is determined primarily by its three‐dimensional structure, the general principles by which structure determines ligand binding remain poorly understood. Here we use statistical analyses of a large number of protein−ligand complexes with associated binding‐affinity measurements to quantitatively characterize how combinations of atomic interactions contribute to ligand affinity. We find that there are significant differences in how atomic interactions determine ligand affinity for proteins that bind small chemical ligands, those that bind DNA/RNA and those that interact with other proteins. Although protein‐small molecule and protein‐DNA/RNA binding affinities can be accurately predicted from structural data, models predicting one type of interaction perform poorly on the others. Additionally, the particular combinations of atomic interactions required to predict binding affinity differed between small‐molecule and DNA/RNA data sets, consistent with the conclusion that the structural bases determining ligand affinity differ among interaction types. In contrast to what we observed for small‐molecule and DNA/RNA interactions, no statistical models were capable of predicting protein−protein affinity with >60% correlation. We demonstrate the potential usefulness of protein‐DNA/RNA binding prediction as a possible tool for high‐throughput virtual screening to guide laboratory investigations, suggesting that quantitative characterization of diverse molecular interactions may have practical applications as well as fundamentally advancing our understanding of how molecular structure translates into function. Proteins 2015; 83:2100–2114. © 2015 The Authors. Proteins: Structure, Function, and Bioinformatics Published by Wiley Periodicals, Inc.

In silico structure-based screening of versatile P-glycoprotein inhibitors using polynomial empirical scoring functions

P-glycoprotein (P-gp) is an ATP (adenosine triphosphate)-binding cassette transporter that causes multidrug resistance of various chemotherapeutic substances by active efflux from mammalian cells. P-gp plays a pivotal role in limiting drug absorption and distribution in different organs, including the intestines and brain. Thus, the prediction of P-gp–drug interactions is of vital importance in assessing drug pharmacokinetic and pharmacodynamic properties. To find the strongest P-gp blockers, we performed an in silico structure-based screening of P-gp inhibitor library (1,300 molecules) by the gradient optimization method, using polynomial empirical scoring (POLSCORE) functions. We report a strong correlation (r²=0.80, F=16.27, n=6, P<0.0157) of inhibition constants (Ki_exp or pKi_exp; experimental Ki or negative decimal logarithm of Ki_exp) converted from experimental IC₅₀ (half maximal inhibitory concentration) values with POLSCORE-predicted constants (Ki_POLSCORE or pKi_POLSCORE), using a linear regression fitting technique. The hydrophobic interactions between P-gp and selected drug substances were detected as the main forces responsible for the inhibition effect. The results showed that this scoring technique might be useful in the virtual screening and filtering of databases of drug-like compounds at the early stage of drug development processes.

Statistical potential for modeling and ranking of protein-ligand interactions

Applications in structural biology and medicinal chemistry require protein-ligand scoring functions for two distinct tasks: (i) ranking different poses of a small molecule in a protein binding site and (ii) ranking different small molecules by their complementarity to a protein site. Using probability theory, we developed two atomic distance-dependent statistical scoring functions: PoseScore was optimized for recognizing native binding geometries of ligands from other poses and RankScore was optimized for distinguishing ligands from nonbinding molecules. Both scores are based on a set of 8,885 crystallographic structures of protein-ligand complexes but differ in the values of three key parameters. Factors influencing the accuracy of scoring were investigated, including the maximal atomic distance and non-native ligand geometries used for scoring, as well as the use of protein models instead of crystallographic structures for training and testing the scoring function. For the test set of 19 targets, RankScore improved the ligand enrichment (logAUC) and early enrichment (EF(1)) scores computed by DOCK 3.6 for 13 and 14 targets, respectively. In addition, RankScore performed better at rescoring than each of seven other scoring functions tested. Accepting both the crystal structure and decoy geometries with all-atom root-mean-square errors of up to 2 Å from the crystal structure as correct binding poses, PoseScore gave the best score to a correct binding pose among 100 decoys for 88% of all cases in a benchmark set containing 100 protein-ligand complexes. PoseScore accuracy is comparable to that of DrugScore(CSD) and ITScore/SE and superior to 12 other tested scoring functions. Therefore, RankScore can facilitate ligand discovery, by ranking complexes of the target with different small molecules; PoseScore can be used for protein-ligand complex structure prediction, by ranking different conformations of a given protein-ligand pair. The statistical potentials are available through the Integrative Modeling Platform (IMP) software package (http://salilab.org/imp) and the LigScore Web server (http://salilab.org/ligscore/).

VoteDock: consensus docking method for prediction of protein-ligand interactions

Molecular recognition plays a fundamental role in all biological processes, and that is why great efforts have been made to understand and predict protein-ligand interactions. Finding a molecule that can potentially bind to a target protein is particularly essential in drug discovery and still remains an expensive and time-consuming task. In silico, tools are frequently used to screen molecular libraries to identify new lead compounds, and if protein structure is known, various protein-ligand docking programs can be used. The aim of docking procedure is to predict correct poses of ligand in the binding site of the protein as well as to score them according to the strength of interaction in a reasonable time frame. The purpose of our studies was to present the novel consensus approach to predict both protein-ligand complex structure and its corresponding binding affinity. Our method used as the input the results from seven docking programs (Surflex, LigandFit, Glide, GOLD, FlexX, eHiTS, and AutoDock) that are widely used for docking of ligands. We evaluated it on the extensive benchmark dataset of 1300 protein-ligands pairs from refined PDBbind database for which the structural and affinity data was available. We compared independently its ability of proper scoring and posing to the previously proposed methods. In most cases, our method is able to dock properly approximately 20% of pairs more than docking methods on average, and over 10% of pairs more than the best single program. The RMSD value of the predicted complex conformation versus its native one is reduced by a factor of 0.5 Å. Finally, we were able to increase the Pearson correlation of the predicted binding affinity in comparison with the experimental value up to 0.5.