One Size Does Not Fit All: The Limits of Structure-Based Models in Drug Discovery

Exploring Scoring Function Space: Developing Computational Models for Drug Discovery

BACKGROUND

The idea of scoring function space established a systems-level approach to address the development of models to predict the affinity of drug molecules by those interested in drug discovery.

OBJECTIVE

Our goal here is to review the concept of scoring function space and how to explore it to develop machine learning models to address protein-ligand binding affinity.

METHODS

We searched the articles available in PubMed related to the scoring function space. We also utilized crystallographic structures found in the protein data bank (PDB) to represent the protein space.

RESULTS

The application of systems-level approaches to address receptor-drug interactions allows us to have a holistic view of the process of drug discovery. The scoring function space adds flexibility to the process since it makes it possible to see drug discovery as a relationship involving mathematical spaces.

CONCLUSION

The application of the concept of scoring function space has provided us with an integrated view of drug discovery methods. This concept is useful during drug discovery, where we see the process as a computational search of the scoring function space to find an adequate model to predict receptor-drug binding affinity.

Sunsetting Binding MOAD with its last data update and the addition of 3D-ligand polypharmacology tools

Binding MOAD is a database of protein-ligand complexes and their affinities with many structured relationships across the dataset. The project has been in development for over 20 years, but now, the time has come to bring it to a close. Currently, the database contains 41,409 structures with affinity coverage for 15,223 (37%) complexes. The website BindingMOAD.org provides numerous tools for polypharmacology exploration. Current relationships include links for structures with sequence similarity, 2D ligand similarity, and binding-site similarity. In this last update, we have added 3D ligand similarity using ROCS to identify ligands which may not necessarily be similar in two dimensions but can occupy the same three-dimensional space. For the 20,387 different ligands present in the database, a total of 1,320,511 3D-shape matches between the ligands were added. Examples of the utility of 3D-shape matching in polypharmacology are presented. Finally, plans for future access to the project data are outlined.

Chemoinformatics and artificial intelligence colloquium: progress and challenges in developing bioactive compounds

We report the main conclusions of the first Chemoinformatics and Artificial Intelligence Colloquium, Mexico City, June 15-17, 2022. Fifteen lectures were presented during a virtual public event with speakers from industry, academia, and non-for-profit organizations. Twelve hundred and ninety students and academics from more than 60 countries. During the meeting, applications, challenges, and opportunities in drug discovery, de novo drug design, ADME-Tox (absorption, distribution, metabolism, excretion and toxicity) property predictions, organic chemistry, peptides, and antibiotic resistance were discussed. The program along with the recordings of all sessions are freely available at https://www.difacquim.com/english/events/2022-colloquium/ .

Novel computational pipelines in antiviral structure‑based drug design (Review)

Viral infections constitute a fundamental and continuous challenge for the global scientific and medical community, as highlighted by the ongoing COVID-19 pandemic. In combination with prophylactic vaccines, the development of safe and effective antiviral drugs remains a pressing need for the effective management of rare and common pathogenic viruses. The design of potent antivirals can be informed by the study of the three-dimensional structure of viral protein targets. Structure-based design of antivirals in silico provides a solution to the arduous and costly process of conventional drug development pipelines. Furthermore, rapid advances in high-throughput computing, along with the growth of available biomolecular and biochemical data, enable the development of novel computational pipelines in the hunt of antivirals. The incorporation of modern methods, such as deep-learning and artificial intelligence, has the potential to revolutionize the structure-based design and repurposing of antiviral compounds, with minimal side effects and high efficacy. The present review aims to provide an outline of both traditional computational drug design and emerging, high-level computing strategies.

Tartgeting Non-alcoholic Fatty Liver Disease: Design, X-Ray co-crystal structure and synthesis of 'first-in-kind' inhibitors of Serine/Threonine Kinase25

We describe the synthesis of a series of 3-t-butyl 5-aminopyrazole p-substituted arylamides as inhibitors of serine-threonine25 (STK25), an enzyme implicated in the progression of non-alcoholic fatty liver disease (NAFLD). Appending a p-N-pyrrolidinosulphonamide group to the arylamide group led to a 'first-in kind' inhibitor with IC₅₀=228nM. A co-crystal structure with STK 25 revealed productive interactions which were also reproduced using molecular docking. A new series of triazolo dihydro oxazine carboxamides of 3-t-butyl 5-aminopyrazole was not active against STK25.

Deciphering the language of antibodies using self-supervised learning

An individual’s B cell receptor (BCR) repertoire encodes information about past immune responses and potential for future disease protection. Deciphering the information stored in BCR sequence datasets will transform our understanding of disease and enable discovery of novel diagnostics and antibody therapeutics. A key challenge of BCR sequence analysis is the prediction of BCR properties from their amino acid sequence alone. Here, we present an antibody-specific language model, Antibody-specific Bidirectional Encoder Representation from Transformers (AntiBERTa), which provides a contextualized representation of BCR sequences. Following pre-training, we show that AntiBERTa embeddings capture biologically relevant information, generalizable to a range of applications. As a case study, we fine-tune AntiBERTa to predict paratope positions from an antibody sequence, outperforming public tools across multiple metrics. To our knowledge, AntiBERTa is the deepest protein-family-specific language model, providing a rich representation of BCRs. AntiBERTa embeddings are primed for multiple downstream tasks and can improve our understanding of the language of antibodies.

•

AntiBERTa is an antibody-specific transformer model for representation learning

•

AntiBERTa embeddings capture aspects of antibody function

•

Attention maps of AntiBERTa correspond to structural contacts and binding sites

•

AntiBERTa can be fine-tuned for state-of-the-art paratope prediction

Understanding antibody function is critical for deciphering the biology of disease and for the discovery of novel therapeutic antibodies. The challenge is the vast diversity of antibody variants compared with the limited labeled data available. We overcome this challenge by using self-supervised learning to train a large antibody-specific language model, followed by transfer learning, to fine-tune the model for predicting information related to antibody function. We initially demonstrate the success of the model by providing leading results in antibody binding site prediction. The model is amenable to further fine-tuning for diverse applications to improve our understanding of antibody function.

Scoring Functions for Protein-Ligand Binding Affinity Prediction using Structure-Based Deep Learning: A Review

The rapid and accurate in silico prediction of protein-ligand binding free energies or binding affinities has the potential to transform drug discovery. In recent years, there has been a rapid growth of interest in deep learning methods for the prediction of protein-ligand binding affinities based on the structural information of protein-ligand complexes. These structure-based scoring functions often obtain better results than classical scoring functions when applied within their applicability domain. Here we review structure-based scoring functions for binding affinity prediction based on deep learning, focussing on different types of architectures, featurization strategies, data sets, methods for training and evaluation, and the role of explainable artificial intelligence in building useful models for real drug-discovery applications.

An in silico pipeline for the discovery of multitarget ligands: A case study for epi-polypharmacology based on DNMT1/HDAC2 inhibition

The search for novel therapeutic compounds remains an overwhelming task owing to the time-consuming and expensive nature of the drug development process and low success rates. Traditional methodologies that rely on the one drug-one target paradigm have proven insufficient for the treatment of multifactorial diseases, leading to a shift to multitarget approaches. In this emerging paradigm, molecules with off-target and promiscuous interactions may result in preferred therapies. In this study, we developed a general pipeline combining machine learning algorithms and a deep generator network to train a dual inhibitor classifier capable of identifying putative pharmacophoric traits. As a case study, we focused on dual inhibitors targeting DNA methyltransferase 1 (DNMT) and histone deacetylase 2 (HDAC2), two enzymes that play a central role in epigenetic regulation. We used this approach to identify dual inhibitors from a novel large natural product database in the public domain. We used docking and atomistic simulations as complementary approaches to establish the ligand-interaction profiles between the best hits and DNMT1/HDAC2. By using the combined ligand- and structure-based approaches, we discovered two promising novel scaffolds that can be used to simultaneously target both DNMT1 and HDAC2. We conclude that the flexibility and adaptability of the proposed pipeline has predictive capabilities of similar or derivative methods and is readily applicable to the discovery of small molecules targeting many other therapeutically relevant proteins.

Learning protein-ligand binding affinity with atomic environment vectors

Scoring functions for the prediction of protein-ligand binding affinity have seen renewed interest in recent years when novel machine learning and deep learning methods started to consistently outperform classical scoring functions. Here we explore the use of atomic environment vectors (AEVs) and feed-forward neural networks, the building blocks of several neural network potentials, for the prediction of protein-ligand binding affinity. The AEV-based scoring function, which we term AEScore, is shown to perform as well or better than other state-of-the-art scoring functions on binding affinity prediction, with an RMSE of 1.22 pK units and a Pearson’s correlation coefficient of 0.83 for the CASF-2016 benchmark. However, AEScore does not perform as well in docking and virtual screening tasks, for which it has not been explicitly trained. Therefore, we show that the model can be combined with the classical scoring function AutoDock Vina in the context of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}Δ-learning, where corrections to the AutoDock Vina scoring function are learned instead of the protein-ligand binding affinity itself. Combined with AutoDock Vina, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}Δ-AEScore has an RMSE of 1.32 pK units and a Pearson’s correlation coefficient of 0.80 on the CASF-2016 benchmark, while retaining the docking and screening power of the underlying classical scoring function.

Machine learning and AI-based approaches for bioactive ligand discovery and GPCR-ligand recognition

In the last decade, machine learning and artificial intelligence applications have received a significant boost in performance and attention in both academic research and industry. The success behind most of the recent state-of-the-art methods can be attributed to the latest developments in deep learning. When applied to various scientific domains that are concerned with the processing of non-tabular data, for example, image or text, deep learning has been shown to outperform not only conventional machine learning but also highly specialized tools developed by domain experts. This review aims to summarize AI-based research for GPCR bioactive ligand discovery with a particular focus on the most recent achievements and research trends. To make this article accessible to a broad audience of computational scientists, we provide instructive explanations of the underlying methodology, including overviews of the most commonly used deep learning architectures and feature representations of molecular data. We highlight the latest AI-based research that has led to the successful discovery of GPCR bioactive ligands. However, an equal focus of this review is on the discussion of machine learning-based technology that has been applied to ligand discovery in general and has the potential to pave the way for successful GPCR bioactive ligand discovery in the future. This review concludes with a brief outlook highlighting the recent research trends in deep learning, such as active learning and semi-supervised learning, which have great potential for advancing bioactive ligand discovery.

Target-Specific Prediction of Ligand Affinity with Structure-Based Interaction Fingerprints

Discovery and optimization of small molecule inhibitors as therapeutic drugs have immensely benefited from rational structure-based drug design. With recent advances in high-resolution structure determination, computational power, and machine learning methodology, it is becoming more tractable to elucidate the structural basis of drug potency. However, the applicability of machine learning models to drug design is limited by the interpretability of the resulting models in terms of feature importance. Here, we take advantage of the large number of available inhibitor-bound HIV-1 protease structures and associated potencies to evaluate inhibitor diversity and machine learning models to predict ligand affinity. First, using a hierarchical clustering approach, we grouped HIV-1 protease inhibitors and identified distinct core structures. Explicit features including protein-ligand interactions were extracted from high-resolution cocrystal structures as 3D-based fingerprints. We found that a gradient boosting machine learning model with this explicit feature attribution can predict binding affinity with high accuracy. Finally, Shapley values were derived to explain local feature importance. We found specific van der Waals (vdW) interactions of key protein residues are pivotal for the predicted potency. Protein-specific and interpretable prediction models can guide the optimization of many small molecule drugs for improved potency.

Predicting Kinase Inhibitor Resistance: Physics-Based and Data-Driven Approaches

Resistance to small molecule drugs often emerges in cancer cells, viruses, and bacteria as a result of the evolutionary pressure exerted by the therapy. Protein mutations that directly impair drug binding are frequently involved in resistance, and the ability to anticipate these mutations would be beneficial in drug development and clinical practice. Here, we evaluate the ability of three distinct computational methods to predict ligand binding affinity changes upon protein mutation for the cancer target Abl kinase. These structure-based approaches rely on first-principle statistical mechanics, mixed physics- and knowledge-based potentials, and machine learning, and were able to estimate binding affinity changes and identify resistant mutations with remarkable accuracy. We expect that these complementary approaches will enable the routine prediction of resistance-causing mutations in a variety of other target proteins.

Unintended specificity of an engineered ligand-binding protein facilitated by unpredicted plasticity of the protein fold

Attempts to create novel ligand-binding proteins often focus on formation of a binding pocket with shape complementarity against the desired ligand (particularly for compounds that lack distinct polar moieties). Although designed proteins often exhibit binding of the desired ligand, in some cases they display unintended recognition behavior. One such designed protein, that was originally intended to bind tetrahydrocannabinol (THC), was found instead to display binding of 25-hydroxy-cholecalciferol (25-D3) and was subjected to biochemical characterization, further selections for enhanced 25-D3 binding affinity and crystallographic analyses. The deviation in specificity is due in part to unexpected altertion of its conformation, corresponding to a significant change of the orientation of an α-helix and an equally large movement of a loop, both of which flank the designed ligand-binding pocket. Those changes led to engineered protein constructs that exhibit significantly more contacts and complementarity towards the 25-D3 ligand than the initial designed protein had been predicted to form towards its intended THC ligand. Molecular dynamics simulations imply that the initial computationally designed mutations may contribute to the movement of the helix. These analyses collectively indicate that accurate prediction and control of backbone dynamics conformation, through a combination of improved conformational sampling and/or de novo structure design, represents a key area of further development for the design and optimization of engineered ligand-binding proteins.

The compromise of virtual screening and its impact on drug discovery

Introduction: Docking and structure-based virtual screening (VS) have been standard approaches in structure-based design for over two decades. However, our understanding of the limitations, potential, and strength of these techniques has enhanced, raising expectations. Areas covered: Based on a survey of reports in the past five years, we assess whether VS: (1) predicts binding poses in agreement with crystallographic data (when available); (2) is a superior screening tool, as often claimed; (3) is successful in identifying chemical scaffolds that can be starting points for subsequent lead optimization cycles. Data shows that knowledge of the target and its chemotypes in postprocessing lead to viable hits in early drug discovery endeavors. Expert opinion: VS is capable of accurate placements in the pocket for the most part, but does not consistently score screening collections accurately. What matters is capitalization on available resources to get closer to a viable lead or optimizable series. Integration of approaches, subjective hit selection guided by knowledge of the receptor or endogenous ligand, libraries driven by experimental guides, validation studies to identify the best docking/scoring that reproduces experimental findings, constraints regarding receptor-ligand interactions, thoroughly designed methodologies, and predefined cutoff scoring criteria strengthen VS's position in pharmaceutical research.

Promiscuous attraction of ligands within the ATP binding site of RyR2 promotes diverse gating behaviour

ATP is an essential constitutive regulator of cardiac ryanodine receptors (RyR2), enabling small changes in cytosolic Ca²⁺ to trigger large changes in channel activity. With recent landmark determinations of the full structures of RyR1 (skeletal isoform) and RyR2 using cryo-EM, and identification of the RyR1 ATP binding site, we have taken the opportunity to model the binding of fragments of ATP into RyR2 in order to investigate how the structure of the ATP site dictates the functional responses of ligands attracted there. RyR2 channel gating was assessed under voltage-clamp conditions and by [³H]ryanodine binding studies. We show that even the triphosphate (PPPi) moiety alone was capable of activating RyR2 but produced two distinct effects (activation or irreversible inactivation) that we suggest correspond to two preferred binding locations within the ATP site. Combinations of complementary fragments of ATP (Pi + ADP or PPi + AMP) could not reproduce the effects of ATP, however, the presence of adenosine prevented the inactivating PPPi effects, allowing activation similar to that of ATP. RyR2 appears to accommodate diverse types of molecules, including PPPi, deep within the ATP binding site. The most effective ligands, however, have at least three phosphate groups that are guided into place by a nucleoside.

SAR-Guided Scoring Function and Mutational Validation Reveal the Binding Mode of CGS-8216 at the α1+/γ2- Benzodiazepine Site

The structural resolution of a bound ligand-receptor complex is a key asset to efficiently drive lead optimization in drug design. However, structural resolution of many drug targets still remains a challenging endeavor. In the absence of structural knowledge, scientists resort to structure-activity relationships (SARs) to promote compound development. In this study, we incorporated ligand-based knowledge to formulate a docking scoring function that evaluates binding poses for their agreement with a known SAR. We showcased this protocol by identifying the binding mode of the pyrazoloquinolinone (PQ) CGS-8216 at the benzodiazepine binding site of the GABA_A receptor. Further evaluation of the final pose by molecular dynamics and free energy simulations revealed a close proximity between the pendent phenyl ring of the PQ and γ2D56, congruent with the low potency of carboxyphenyl analogues. Ultimately, we introduced the γ2D56A mutation and in fact observed a 10-fold potency increase in the carboxyphenyl analogue, providing experimental evidence in favor of our binding hypothesis.

Predictive Power of Different Types of Experimental Restraints in Small Molecule Docking: A Review

Incorporating experimental restraints is a powerful method of increasing accuracy in computational protein small molecule docking simulations. Different algorithms integrate distinct forms of biochemical data during the docking and/or scoring stages. These so-called hybrid methods make use of receptor-based information such as nuclear magnetic resonance (NMR) restraints or small molecule-based information such as structure-activity relationships (SARs). A third class of methods directly interrogates contacts between the protein receptor and the small molecule. This work reviews the current state of using such restraints in docking simulations, evaluates their feasibility across broad systems, and identifies potential areas of algorithm development.

Antibody side chain conformations are position-dependent

Side chain prediction is an integral component of computational antibody design and structure prediction. Current antibody modelling tools use backbone-dependent rotamer libraries with conformations taken from general proteins. Here we present our antibody-specific rotamer library, where rotamers are binned according to their immunogenetics (IMGT) position, rather than their local backbone geometry. We find that for some amino acid types at certain positions, only a restricted number of side chain conformations are ever observed. Using this information, we are able to reduce the breadth of the rotamer sampling space. Based on our rotamer library, we built a side chain predictor, position-dependent antibody rotamer swapper (PEARS). On a blind test set of 95 antibody model structures, PEARS had the highest average χ₁ and χ1+2 accuracy (78.7% and 64.8%) compared to three leading backbone-dependent side chain predictors. Our use of IMGT position, rather than backbone ϕ/ψ, meant that PEARS was more robust to errors in the backbone of the model structure. PEARS also achieved the lowest number of side chain-side chain clashes. PEARS is freely available as a web application at http://opig.stats.ox.ac.uk/webapps/pears.

Thienopyrimidinone Based Sirtuin-2 (SIRT2)-Selective Inhibitors Bind in the Ligand Induced Selectivity Pocket

Sirtuins (SIRTs) are NAD-dependent deacylases, known to be involved in a variety of pathophysiological processes and thus remain promising therapeutic targets for further validation. Previously, we reported a novel thienopyrimidinone SIRT2 inhibitor with good potency and excellent selectivity for SIRT2. Herein, we report an extensive SAR study of this chemical series and identify the key pharmacophoric elements and physiochemical properties that underpin the excellent activity observed. New analogues have been identified with submicromolar SIRT2 inhibtory activity and good to excellent SIRT2 subtype-selectivity. Importantly, we report a cocrystal structure of one of our compounds (29c) bound to SIRT2. This reveals our series to induce the formation of a previously reported selectivity pocket but to bind in an inverted fashion to what might be intuitively expected. We believe these findings will contribute significantly to an understanding of the mechanism of action of SIRT2 inhibitors and to the identification of refined, second generation inhibitors.

Accurate Prediction of Ligand Affinities for a Proton-Dependent Oligopeptide Transporter

Membrane transporters are critical modulators of drug pharmacokinetics, efficacy, and safety. One example is the proton-dependent oligopeptide transporter PepT1, also known as SLC15A1, which is responsible for the uptake of the β-lactam antibiotics and various peptide-based prodrugs. In this study, we modeled the binding of various peptides to a bacterial homolog, PepT_St, and evaluated a range of computational methods for predicting the free energy of binding. Our results show that a hybrid approach (endpoint methods to classify peptides into good and poor binders and a theoretically exact method for refinement) is able to accurately predict affinities, which we validated using proteoliposome transport assays. Applying the method to a homology model of PepT1 suggests that the approach requires a high-quality structure to be accurate. Our study provides a blueprint for extending these computational methodologies to other pharmaceutically important transporter families.

•

A hierarchical computational approach determines ligand affinities to transporters

•

Lysine-containing dipeptides proposed to bind vertically like a tripeptide

•

Experimental structures are vital for the accurate prediction of affinities

•

A model of prodrug interactions to human PepT1 is suggested

Accurate calculation of the absolute free energy of binding for drug molecules

Accurate prediction of binding affinities has been a central goal of computational chemistry for decades, yet remains elusive. Despite good progress, the required accuracy for use in a drug-discovery context has not been consistently achieved for drug-like molecules. Here, we perform absolute free energy calculations based on a thermodynamic cycle for a set of diverse inhibitors binding to bromodomain-containing protein 4 (BRD4) and demonstrate that a mean absolute error of 0.6 kcal mol-1 can be achieved. We also show a similar level of accuracy (1.0 kcal mol-1) can be achieved in pseudo prospective approach. Bromodomains are epigenetic mark readers that recognize acetylation motifs and regulate gene transcription, and are currently being investigated as therapeutic targets for cancer and inflammation. The unprecedented accuracy offers the exciting prospect that the binding free energy of drug-like compounds can be predicted for pharmacologically relevant targets.

Progress and challenges in predicting protein interfaces

The majority of biological processes are mediated via protein-protein interactions. Determination of residues participating in such interactions improves our understanding of molecular mechanisms and facilitates the development of therapeutics. Experimental approaches to identifying interacting residues, such as mutagenesis, are costly and time-consuming and thus, computational methods for this purpose could streamline conventional pipelines. Here we review the field of computational protein interface prediction. We make a distinction between methods which address proteins in general and those targeted at antibodies, owing to the radically different binding mechanism of antibodies. We organize the multitude of currently available methods hierarchically based on required input and prediction principles to provide an overview of the field.

Beyond Membrane Protein Structure: Drug Discovery, Dynamics and Difficulties

Most of the previous content of this book has focused on obtaining the structures of membrane proteins. In this chapter we explore how those structures can be further used in two key ways. The first is their use in structure based drug design (SBDD) and the second is how they can be used to extend our understanding of their functional activity via the use of molecular dynamics. Both aspects now heavily rely on computations. This area is vast, and alas, too large to consider in depth in a single book chapter. Thus where appropriate we have referred the reader to recent reviews for deeper assessment of the field. We discuss progress via the use of examples from two main drug target areas; G-protein coupled receptors (GPCRs) and ion channels. We end with a discussion of some of the main challenges in the area.

Charting a Path to Success in Virtual Screening

Docking is commonly applied to drug design efforts, especially high-throughput virtual screenings of small molecules, to identify new compounds that bind to a given target. Despite great advances and successful applications in recent years, a number of issues remain unsolved. Most of the challenges and problems faced when running docking experiments are independent of the specific software used, and can be ascribed to either improper input preparation or to the simplified approaches applied to achieve high-throughput speed. Being aware of approximations and limitations of such methods is essential to prevent errors, deal with misleading results, and increase the success rate of virtual screening campaigns. In this review, best practices and most common issues of docking and virtual screening will be discussed, covering the journey from the design of the virtual experiment to the hit identification.

Improvements, trends, and new ideas in molecular docking: 2012-2013 in review

Molecular docking is a computational method for predicting the placement of ligands in the binding sites of their receptor(s). In this review, we discuss the methodological developments that occurred in the docking field in 2012 and 2013, with a particular focus on the more difficult aspects of this computational discipline. The main challenges and therefore focal points for developments in docking, covered in this review, are receptor flexibility, solvation, scoring, and virtual screening. We specifically deal with such aspects of molecular docking and its applications as selection criteria for constructing receptor ensembles, target dependence of scoring functions, integration of higher-level theory into scoring, implicit and explicit handling of solvation in the binding process, and comparison and evaluation of docking and scoring methods.

Computational studies to predict or explain G protein coupled receptor polypharmacology

Since G protein-coupled receptors (GPCRs) belong to a very large superfamily of evolutionarily related receptors (>800 members in humans), and due to the rapid progress on their structural biology, they are ideal candidates for polypharmacology studies. Broad screening and bioinformatics/chemoinformatics have been applied to understanding off-target effects of GPCR ligands. It is now feasible to approach the question of GPCR polypharmacology using molecular modeling and the available X-ray GPCR structures. As an example, large and sterically constrained adenosine derivatives (potent adenosine receptor ligands with low conformational freedom and multiple extended substituents) were screened for binding at diverse receptors. Unanticipated off-target interactions, including at biogenic amine receptors, were then modeled using a structure-based approach to provide a consistent understanding of recognition. A conserved Asp in TM3 changed its role from counterion for biogenic amines to characteristic H-bonding to adenosine. The same systematic approach could potentially be applied to many GPCRs or other receptors using other sets of congeneric ligands.

Substituting random forest for multiple linear regression improves binding affinity prediction of scoring functions: Cyscore as a case study

BACKGROUND

State-of-the-art protein-ligand docking methods are generally limited by the traditionally low accuracy of their scoring functions, which are used to predict binding affinity and thus vital for discriminating between active and inactive compounds. Despite intensive research over the years, classical scoring functions have reached a plateau in their predictive performance. These assume a predetermined additive functional form for some sophisticated numerical features, and use standard multivariate linear regression (MLR) on experimental data to derive the coefficients.

RESULTS

In this study we show that such a simple functional form is detrimental for the prediction performance of a scoring function, and replacing linear regression by machine learning techniques like random forest (RF) can improve prediction performance. We investigate the conditions of applying RF under various contexts and find that given sufficient training samples RF manages to comprehensively capture the non-linearity between structural features and measured binding affinities. Incorporating more structural features and training with more samples can both boost RF performance. In addition, we analyze the importance of structural features to binding affinity prediction using the RF variable importance tool. Lastly, we use Cyscore, a top performing empirical scoring function, as a baseline for comparison study.

CONCLUSIONS

Machine-learning scoring functions are fundamentally different from classical scoring functions because the former circumvents the fixed functional form relating structural features with binding affinities. RF, but not MLR, can effectively exploit more structural features and more training samples, leading to higher prediction performance. The future availability of more X-ray crystal structures will further widen the performance gap between RF-based and MLR-based scoring functions. This further stresses the importance of substituting RF for MLR in scoring function development.