Towards automated binding affinity prediction using an iterative linear interaction energy approach

Molecular Dynamics as a Tool for Virtual Ligand Screening

Rational drug design is essential for new drugs to emerge, especially when the structure of a target protein or nucleic acid is known. To that purpose, high-throughput virtual ligand screening campaigns aim at discovering computationally new binding molecules or fragments to modulate particular biomolecular interactions or biological activities, related to a disease process. The structure-based virtual ligand screening process primarily relies on docking methods which allow predicting the binding of a molecule to a biological target structure with a correct conformation and the best possible affinity. The docking method itself is not sufficient as it suffers from several and crucial limitations (lack of full protein flexibility information, no solvation and ion effects, poor scoring functions, and unreliable molecular affinity estimation).At the interface of computer techniques and drug discovery, molecular dynamics (MD) allows introducing protein flexibility before or after a docking protocol, refining the structure of protein-drug complexes in the presence of water, ions, and even in membrane-like environments, describing more precisely the temporal evolution of the biological complex and ranking these complexes with more accurate binding energy calculations. In this chapter, we describe the up-to-date MD, which plays the role of supporting tools in the virtual ligand screening (VS) process.Without a doubt, using docking in combination with MD is an attractive approach in structure-based drug discovery protocols nowadays. It has proved its efficiency through many examples in the literature and is a powerful method to significantly reduce the amount of required wet experimentations (Tarcsay et al, J Chem Inf Model 53:2990-2999, 2013; Barakat et al, PLoS One 7:e51329, 2012; De Vivo et al, J Med Chem 59:4035-4061, 2016; Durrant, McCammon, BMC Biol 9:71-79, 2011; Galeazzi, Curr Comput Aided Drug Des 5:225-240, 2009; Hospital et al, Adv Appl Bioinforma Chem 8:37-47, 2015; Jiang et al, Molecules 20:12769-12786, 2015; Kundu et al, J Mol Graph Model 61:160-174, 2015; Mirza et al, J Mol Graph Model 66:99-107, 2016; Moroy et al, Future Med Chem 7:2317-2331, 2015; Naresh et al, J Mol Graph Model 61:272-280, 2015; Nichols et al, J Chem Inf Model 51:1439-1446, 2011; Nichols et al, Methods Mol Biol 819:93-103, 2012; Okimoto et al, PLoS Comput Biol 5:e1000528, 2009; Rodriguez-Bussey et al, Biopolymers 105:35-42, 2016; Sliwoski et al, Pharmacol Rev 66:334-395, 2014).

Structural bioinformatics analysis of SARS-CoV-2 variants reveals higher hACE2 receptor binding affinity for Omicron B.1.1.529 spike RBD compared to wild type reference

To date, more than 263 million people have been infected with SARS-CoV-2 during the COVID-19 pandemic. In many countries, the global spread occurred in multiple pandemic waves characterized by the emergence of new SARS-CoV-2 variants. Here we report a sequence and structural-bioinformatics analysis to estimate the effects of amino acid substitutions on the affinity of the SARS-CoV-2 spike receptor binding domain (RBD) to the human receptor hACE2. This is done through qualitative electrostatics and hydrophobicity analysis as well as molecular dynamics simulations used to develop a high-precision empirical scoring function (ESF) closely related to the linear interaction energy method and calibrated on a large set of experimental binding energies. For the latest variant of concern (VOC), B.1.1.529 Omicron, our Halo difference point cloud studies reveal the largest impact on the RBD binding interface compared to all other VOC. Moreover, according to our ESF model, Omicron achieves a much higher ACE2 binding affinity than the wild type and, in particular, the highest among all VOCs except Alpha and thus requires special attention and monitoring.

Structural systems pharmacology: A framework for integrating metabolic network and structure-based virtual screening for drug discovery against bacteria

Advances in genome-scale metabolic models (GEMs) and computational drug discovery have caused the identification of drug targets at the system-level and inhibitors to combat bacterial infection and drug resistance. Here we report a structural systems pharmacology framework that integrates the GEM and structure-based virtual screening (SBVS) method to identify drugs effective for Escherichia coli infection. The most complete genome-scale metabolic reconstruction integrated with protein structures (GEM-PRO) of E. coli, iML1515_GP, and FDA-approved drugs have been used. FBA was performed to predict drug targets in silico. The 195 essential genes were predicted in the rich medium. The subsystems in which a significant number of these genes are involved are cofactor, lipopolysaccharide (LPS) biosynthesis that are necessary for cell growth. Therefore, some proteins encoded by these genes are responsible for the biosynthesis and transport of LPS which is the first line of defense against threats. So, these proteins can be potential drug targets. The enzymes with experimental structure and cognate ligands were selected as final drug targets for performing the SBVS method. Finally, we have suggested those drugs that have good interaction with the selected proteins as drug repositioning cases. Also, the suggested molecules could be promising lead compounds. This framework may be helpful to fill the gap between genomics and drug discovery. Results show this framework suggests novel antibacterials that can be subjected to experimental testing soon and it can be suitable for other pathogens.

Computing Cellulase Kinetics with a Two-Domain Linear Interaction Energy Approach

While heterogeneous enzyme reactions play an essential role in both nature and green industries, computational predictions of their catalytic properties remain scarce. Recent experimental work demonstrated the applicability of the Sabatier principle for heterogeneous biocatalysis. This provides a simple relationship between binding strength and the catalytic rate and potentially opens a new way for inexpensive computational determination of kinetic parameters. However, broader implementation of this approach will require fast and reliable prediction of binding free energies of complex two-phase systems, and computational procedures for this are still elusive. Here, we propose a new framework for the assessment of the binding strengths of multidomain proteins, in general, and interfacial enzymes, in particular, based on an extended linear interaction energy (LIE) method. This two-domain LIE (2D-LIE) approach was successfully applied to predict binding and activation free energies of a diverse set of cellulases and resulted in robust models with high accuracy. Overall, our method provides a fast computational screening tool for cellulases that have not been experimentally characterized, and we posit that it may also be applicable to other heterogeneously acting biocatalysts.

How Good is Jarzynski's Equality for Computer-Aided Drug Design?

Accurate determination of the binding affinity of the ligand to the receptor remains a difficult problem in computer-aided drug design. Here, we study and compare the efficiency of Jarzynski’s equality (JE) combined with steered molecular dynamics and the linear interaction energy (LIE) method by assessing the binding affinity of 23 small compounds to six receptors, including β-lactamase, thrombin, factor Xa, HIV-1 protease (HIV), myeloid cell leukemia-1, and cyclin-dependent kinase 2 proteins. It was shown that Jarzynski’s nonequilibrium binding free energy ΔG_neq^Jar correlates with the available experimental data with the correlation levels R = 0.89, 0.86, 0.83, 0.80, 0.83, and 0.81 for six data sets, while for the binding free energy ΔG_LIE obtained by the LIE method, we have R = 0.73, 0.80, 0.42, 0.23, 0.85, and 0.01. Therefore, JE is recommended to be used for ranking binding affinities as it provides accurate and robust results. In contrast, LIE is not as reliable as JE, and it should be used with caution, especially when it comes to new systems.

Recent Developments in Linear Interaction Energy Based Binding Free Energy Calculations

The linear interaction energy (LIE) approach is an end-point method to compute binding affinities. As such it combines explicit conformational sampling (of the protein-bound and unbound-ligand states) with efficiency in calculating values for the protein-ligand binding free energy ΔG _bind . This perspective summarizes our recent efforts to use molecular simulation and empirically calibrated LIE models for accurate and efficient calculation of ΔG _bind for diverse sets of compounds binding to flexible proteins (e.g., Cytochrome P450s and other proteins of direct pharmaceutical or biochemical interest). Such proteins pose challenges on ΔG _bind computation, which we tackle using a previously introduced statistically weighted LIE scheme. Because calibrated LIE models require empirical fitting of scaling parameters, they need to be accompanied with an applicability domain (AD) definition to provide a measure of confidence for predictions for arbitrary query compounds within a reference frame defined by a collective chemical and interaction space. To enable AD assessment of LIE predictions (or other protein-structure and -dynamic based ΔG _bind calculations) we recently introduced strategies for AD assignment of LIE models, based on simulation and training data only. These strategies are reviewed here as well, together with available tools to facilitate and/or automate LIE computation (including software for combined statistically-weighted LIE calculations and AD assessment).

Optimal designs for pairwise calculation: An application to free energy perturbation in minimizing prediction variability

Pairwise-based methods such as the free energy perturbation (FEP) method have been widely deployed to compute the binding free energy differences between two similar host-guest complexes. The calculated pairwise free energy difference is either directly adopted or transformed to absolute binding free energy for molecule rank ordering. We investigated, through both analytic derivations and simulations, how the selection of pairs in the experiment could impact the overall prediction precision. Our studies showed that (1) the estimated absolute binding free energy ( Δ G ^ ) derived from calculated pairwise differences (ΔΔG) through weighted least squares fitting is more precise in prediction than the pairwise difference values when the number of pairs is more than the number of ligands and (2) prediction precision is influenced by both the total number of pairs and the specifically selected pairs, the latter being critically important when the number of calculated pairs is limited. Furthermore, we applied optimal experimental design in pair selection and found that the optimally selected pairs can outperform randomly selected pairs in prediction precision. In an illustrative example, we showed that, upon weighing ligand structure similarity into design optimization, the weighted optimal designs are more efficient than the literature reported designs. This work provides a new approach to assess retrospective pairwise-based prediction results, and a method to design new prospective pairwise-based experiments for molecular lead optimization. © 2019 Wiley Periodicals, Inc.

A Comparative Linear Interaction Energy and MM/PBSA Study on SIRT1-Ligand Binding Free Energy Calculation

Binding free energy (ΔGbind) computation can play an important role in prioritizing compounds to be evaluated experimentally on their affinity for target proteins, yet fast and accurate ΔGbind calculation remains an elusive task. In this study, we compare the performance of two popular end-point methods, i.e., linear interaction energy (LIE) and molecular mechanics/Poisson-Boltzmann surface area (MM/PBSA), with respect to their ability to correlate calculated binding affinities of 27 thieno[3,2-d]pyrimidine-6-carboxamide-derived sirtuin 1 (SIRT1) inhibitors with experimental data. Compared with the standard single-trajectory setup of MM/PBSA, our study elucidates that LIE allows to obtain direct ("absolute") values for SIRT1 binding free energies with lower compute requirements, while the accuracy in calculating relative values for ΔGbind is comparable (Pearson's r = 0.72 and 0.64 for LIE and MM/PBSA, respectively). We also investigate the potential of combining multiple docking poses in iterative LIE models and find that Boltzmann-like weighting of outcomes of simulations starting from different poses can retrieve appropriate binding orientations. In addition, we find that in this particular case study the LIE and MM/PBSA models can be optimized by neglecting the contributions from electrostatic and polar interactions to the ΔGbind calculations.

Probing inhibition mechanisms of adenosine deaminase by using molecular dynamics simulations

Adenosine deaminase (ADA) catalyzes the deamination of adenosine, which is important in purine metabolism. ADA is ubiquitous to almost all human tissues, and ADA abnormalities have been reported in various diseases, including rheumatoid arthritis. ADA can be divided into two conformations based on the inhibitor that it binds to: open and closed forms. Here, we chose three ligands, namely, FR117016 (FR0), FR221647 (FR2) (open form), and HDPR (PRH, closed form), to investigate the inhibition mechanism of ADA and its effect on ADA through molecular dynamics simulations. In open forms, Egap and electrostatic potential (ESP) indicated that electron transfer might occur more easily in FR0 than in FR2. Binding free energy and hydrogen bond occupation revealed that the ADA-FR0 complex had a more stable structure than ADA-FR2. The probability of residues Pro159 to Lys171 of ADA-FR0 and ADA-FR2 to form a helix moderately increased compared with that in nonligated ADA. In comparison with FR0 and FR2 PRH could maintain ADA in a closed form to inhibit the function of ADA. The α7 helix (residues Thr57 to Ala73) of ADA in the closed form was mostly unfastened because of the effect of PRH. The number of H bonds and the relative superiority of the binding free energy indicated that the binding strength of PRH to ADA was significantly lower than that of an open inhibitor, thereby supporting the comparison of the inhibitory activities of the three ligands. Alanine scanning results showed that His17, Gly184, Asp295, and Asp296 exerted the greatest effects on protein energy, suggesting that they played crucial roles in binding to inhibitors. This study served as a theoretical basis for the development of new ADA inhibitors.

Calculate protein-ligand binding affinities with the extended linear interaction energy method: application on the Cathepsin S set in the D3R Grand Challenge 3

We participated in the Cathepsin S (CatS) sub-challenge of the Drug Design Data Resource (D3R) Grand Challenge 3 (GC3) in 2017 to blindly predict the binding poses of 24 CatS-bound ligands, the binding affinity ranking of 136 ligands, and the binding free energies of a subset of 33 ligands in Stage 1A and Stage 2. Our submitted predictions ranked relatively well compared to the submissions from other participants. Here we present our methodologies used in the challenge. For the binding pose prediction, we employed the Glide module in the Schrodinger Suite 2017 and AutoDock Vina. For the binding affinity/free energy prediction, we carried out molecular dynamics simulations of the complexes in explicit water solvent with counter ions, and then estimated the binding free energies with our newly developed model of extended linear interaction energy (ELIE), which is inspired by two other popular end-point approaches: the linear interaction energy (LIE) method, and the molecular mechanics with Poisson-Boltzmann surface area solvation method (MM/PBSA). Our studies suggest that ELIE is a good trade-off between efficiency and accuracy, and it is appropriate for filling the gap between the high-throughput docking and scoring methods and the rigorous but much more computationally demanding methods like free energy perturbation (FEP) or thermodynamics integration (TI) in computer-aided drug design (CADD) projects.

Molecular Dynamics as a Tool for Virtual Ligand Screening

Rational drug design is essential for new drugs to emerge, especially when the structure of a target protein or catalytic enzyme is known experimentally. To that purpose, high-throughput virtual ligand screening campaigns aim at discovering computationally new binding molecules or fragments to inhibit a particular protein interaction or biological activity. The virtual ligand screening process often relies on docking methods which allow predicting the binding of a molecule into a biological target structure with a correct conformation and the best possible affinity. The docking method itself is not sufficient as it suffers from several and crucial limitations (lack of protein flexibility information, no solvation effects, poor scoring functions, and unreliable molecular affinity estimation).At the interface of computer techniques and drug discovery, molecular dynamics (MD) allows introducing protein flexibility before or after a docking protocol, refining the structure of protein-drug complexes in the presence of water, ions and even in membrane-like environments, and ranking complexes with more accurate binding energy calculations. In this chapter we describe the up-to-date MD protocols that are mandatory supporting tools in the virtual ligand screening (VS) process. Using docking in combination with MD is one of the best computer-aided drug design protocols nowadays. It has proved its efficiency through many examples, described below.

Binding free energy predictions of farnesoid X receptor (FXR) agonists using a linear interaction energy (LIE) approach with reliability estimation: application to the D3R Grand Challenge 2

Computational protein binding affinity prediction can play an important role in drug research but performing efficient and accurate binding free energy calculations is still challenging. In the context of phase 2 of the Drug Design Data Resource (D3R) Grand Challenge 2 we used our automated eTOX ALLIES approach to apply the (iterative) linear interaction energy (LIE) method and we evaluated its performance in predicting binding affinities for farnesoid X receptor (FXR) agonists. Efficiency was obtained by our pre-calibrated LIE models and molecular dynamics (MD) simulations at the nanosecond scale, while predictive accuracy was obtained for a small subset of compounds. Using our recently introduced reliability estimation metrics, we could classify predictions with higher confidence by featuring an applicability domain (AD) analysis in combination with protein-ligand interaction profiling. The outcomes of and agreement between our AD and interaction-profile analyses to distinguish and rationalize the performance of our predictions highlighted the relevance of sufficiently exploring protein-ligand interactions during training and it demonstrated the possibility to quantitatively and efficiently evaluate if this is achieved by using simulation data only.

eTOX ALLIES: an automated pipeLine for linear interaction energy-based simulations

BACKGROUND

Computational methods to predict binding affinities of small ligands toward relevant biological (off-)targets are helpful in prioritizing the screening and synthesis of new drug candidates, thereby speeding up the drug discovery process. However, use of ligand-based approaches can lead to erroneous predictions when structural and dynamic features of the target substantially affect ligand binding. Free energy methods for affinity computation can include steric and electrostatic protein-ligand interactions, solvent effects, and thermal fluctuations, but often they are computationally demanding and require a high level of supervision. As a result their application is typically limited to the screening of small sets of compounds by experts in molecular modeling.

RESULTS

We have developed eTOX ALLIES, an open source framework that allows the automated prediction of ligand-binding free energies requiring the ligand structure as only input. eTOX ALLIES is based on the linear interaction energy approach, an efficient end-point free energy method derived from Free Energy Perturbation theory. Upon submission of a ligand or dataset of compounds, the tool performs the multiple steps required for binding free-energy prediction (docking, ligand topology creation, molecular dynamics simulations, data analysis), making use of external open source software where necessary. Moreover, functionalities are also available to enable and assist the creation and calibration of new models. In addition, a web graphical user interface has been developed to allow use of free-energy based models to users that are not an expert in molecular modeling.

CONCLUSIONS

Because of the user-friendliness, efficiency and free-software licensing, eTOX ALLIES represents a novel extension of the toolbox for computational chemists, pharmaceutical scientists and toxicologists, who are interested in fast affinity predictions of small molecules toward biological (off-)targets for which protein flexibility, solvent and binding site interactions directly affect the strength of ligand-protein binding.

Comprehensive and Automated Linear Interaction Energy Based Binding-Affinity Prediction for Multifarious Cytochrome P450 Aromatase Inhibitors

Cytochrome P450 aromatase (CYP19A1) plays a key role in the development of estrogen dependent breast cancer, and aromatase inhibitors have been at the front line of treatment for the past three decades. The development of potent, selective and safer inhibitors is ongoing with in silico screening methods playing a more prominent role in the search for promising lead compounds in bioactivity-relevant chemical space. Here we present a set of comprehensive binding affinity prediction models for CYP19A1 using our automated Linear Interaction Energy (LIE) based workflow on a set of 132 putative and structurally diverse aromatase inhibitors obtained from a typical industrial screening study. We extended the workflow with machine learning methods to automatically cluster training and test compounds in order to maximize the number of explained compounds in one or more predictive LIE models. The method uses protein-ligand interaction profiles obtained from Molecular Dynamics (MD) trajectories to help model search and define the applicability domain of the resolved models. Our method was successful in accounting for 86% of the data set in 3 robust models that show high correlation between calculated and observed values for ligand-binding free energies (RMSE < 2.5 kJ mol-1), with good cross-validation statistics.

PL-PatchSurfer2: Improved Local Surface Matching-Based Virtual Screening Method That Is Tolerant to Target and Ligand Structure Variation

Virtual screening has become an indispensable procedure in drug discovery. Virtual screening methods can be classified into two categories: ligand-based and structure-based. While the former have advantages, including being quick to compute, in general they are relatively weak at discovering novel active compounds because they use known actives as references. On the other hand, structure-based methods have higher potential to find novel compounds because they directly predict the binding affinity of a ligand in a target binding pocket, albeit with substantially lower speed than ligand-based methods. Here we report a novel structure-based virtual screening method, PL-PatchSurfer2. In PL-PatchSurfer2, protein and ligand surfaces are represented by a set of overlapping local patches, each of which is represented by three-dimensional Zernike descriptors (3DZDs). By means of 3DZDs, the shapes and physicochemical complementarities of local surface regions of a pocket surface and a ligand molecule can be concisely and effectively computed. Compared with the previous version of the program, the performance of PL-PatchSurfer2 is substantially improved by the addition of two more features, atom-based hydrophobicity and hydrogen-bond acceptors and donors. Benchmark studies showed that PL-PatchSurfer2 performed better than or comparable to popular existing methods. Particularly, PL-PatchSurfer2 significantly outperformed existing methods when apo-form or template-based protein models were used for queries. The computational time of PL-PatchSurfer2 is about 20 times shorter than those of conventional structure-based methods. The PL-PatchSurfer2 program is available at http://www.kiharalab.org/plps2/ .

Improving the iterative Linear Interaction Energy approach using automated recognition of configurational transitions

Recently an iterative method was proposed to enhance the accuracy and efficiency of ligand-protein binding affinity prediction through linear interaction energy (LIE) theory. For ligand binding to flexible Cytochrome P450s (CYPs), this method was shown to decrease the root-mean-square error and standard deviation of error prediction by combining interaction energies of simulations starting from different conformations. Thereby, different parts of protein-ligand conformational space are sampled in parallel simulations. The iterative LIE framework relies on the assumption that separate simulations explore different local parts of phase space, and do not show transitions to other parts of configurational space that are already covered in parallel simulations. In this work, a method is proposed to (automatically) detect such transitions during the simulations that are performed to construct LIE models and to predict binding affinities. Using noise-canceling techniques and splines to fit time series of the raw data for the interaction energies, transitions during simulation between different parts of phase space are identified. Boolean selection criteria are then applied to determine which parts of the interaction energy trajectories are to be used as input for the LIE calculations. Here we show that this filtering approach benefits the predictive quality of our previous CYP 2D6-aryloxypropanolamine LIE model. In addition, an analysis is performed of the gain in computational efficiency that can be obtained from monitoring simulations using the proposed filtering method and by prematurely terminating simulations accordingly.

Linear Interaction Energy Based Prediction of Cytochrome P450 1A2 Binding Affinities with Reliability Estimation

Prediction of human Cytochrome P450 (CYP) binding affinities of small ligands, i.e., substrates and inhibitors, represents an important task for predicting drug-drug interactions. A quantitative assessment of the ligand binding affinity towards different CYPs can provide an estimate of inhibitory activity or an indication of isoforms prone to interact with the substrate of inhibitors. However, the accuracy of global quantitative models for CYP substrate binding or inhibition based on traditional molecular descriptors can be limited, because of the lack of information on the structure and flexibility of the catalytic site of CYPs. Here we describe the application of a method that combines protein-ligand docking, Molecular Dynamics (MD) simulations and Linear Interaction Energy (LIE) theory, to allow for quantitative CYP affinity prediction. Using this combined approach, a LIE model for human CYP 1A2 was developed and evaluated, based on a structurally diverse dataset for which the estimated experimental uncertainty was 3.3 kJ mol-1. For the computed CYP 1A2 binding affinities, the model showed a root mean square error (RMSE) of 4.1 kJ mol-1 and a standard error in prediction (SDEP) in cross-validation of 4.3 kJ mol-1. A novel approach that includes information on both structural ligand description and protein-ligand interaction was developed for estimating the reliability of predictions, and was able to identify compounds from an external test set with a SDEP for the predicted affinities of 4.6 kJ mol-1 (corresponding to 0.8 pKi units).

The eTOX data-sharing project to advance in silico drug-induced toxicity prediction

The high-quality in vivo preclinical safety data produced by the pharmaceutical industry during drug development, which follows numerous strict guidelines, are mostly not available in the public domain. These safety data are sometimes published as a condensed summary for the few compounds that reach the market, but the majority of studies are never made public and are often difficult to access in an automated way, even sometimes within the owning company itself. It is evident from many academic and industrial examples, that useful data mining and model development requires large and representative data sets and careful curation of the collected data. In 2010, under the auspices of the Innovative Medicines Initiative, the eTOX project started with the objective of extracting and sharing preclinical study data from paper or pdf archives of toxicology departments of the 13 participating pharmaceutical companies and using such data for establishing a detailed, well-curated database, which could then serve as source for read-across approaches (early assessment of the potential toxicity of a drug candidate by comparison of similar structure and/or effects) and training of predictive models. The paper describes the efforts undertaken to allow effective data sharing intellectual property (IP) protection and set up of adequate controlled vocabularies) and to establish the database (currently with over 4000 studies contributed by the pharma companies corresponding to more than 1400 compounds). In addition, the status of predictive models building and some specific features of the eTOX predictive system (eTOXsys) are presented as decision support knowledge-based tools for drug development process at an early stage.