Extended connectivity interaction features: improving binding affinity prediction through chemical description

Harnessing pre-trained models for accurate prediction of protein-ligand binding affinity

BACKGROUND

The binding between proteins and ligands plays a crucial role in the field of drug discovery. However, this area currently faces numerous challenges. On one hand, existing methods are constrained by the limited availability of labeled data, often performing inadequately when addressing complex protein-ligand interactions. On the other hand, many models struggle to effectively capture the flexible variations and relative spatial relationships between proteins and ligands. These issues not only significantly hinder the advancement of protein-ligand binding research but also adversely affect the accuracy and efficiency of drug discovery. Therefore, in response to these challenges, our study aims to enhance predictive capabilities through innovative approaches, providing more reliable support for drug discovery efforts.

METHODS

This study leverages a pre-trained model with spatial awareness to enhance the prediction of protein-ligand binding affinity. By perturbing the structures of small molecules in a manner consistent with physical constraints and employing self-supervised tasks, we improve the representation of small molecule structures, allowing for better adaptation to affinity predictions. Meanwhile, our approach enables the identification of potential binding sites on proteins.

RESULTS

Our model demonstrates a significantly higher correlation coefficient in binding affinity predictions. Extensive evaluation on the PDBBind v2019 refined set, CASF, and Merck FEP benchmarks confirms the model's robustness and strong generalization across diverse datasets. Additionally, the model achieves over 95% in classification ROC for binding site identification, underscoring its high accuracy in pinpointing protein-ligand interaction regions.

CONCLUSION

This research presents a novel approach that not only enhances the accuracy of binding affinity predictions but also facilitates the identification of binding sites, showcasing the potential of pre-trained models in computational drug design. Data and code are available at https://github.com/MIALAB-RUC/SableBind .

Narrowing the gap between machine learning scoring functions and free energy perturbation using augmented data

Machine learning offers great promise for fast and accurate binding affinity predictions. However, current models lack robust evaluation and fail on tasks encountered in (hit-to-) lead optimisation, such as ranking the binding affinity of a congeneric series of ligands, thereby limiting their application in drug discovery. Here, we address these issues by first introducing a novel attention-based graph neural network model called AEV-PLIG (atomic environment vector-protein ligand interaction graph). Second, we introduce a new and more realistic out-of-distribution test set called the OOD Test. We benchmark our model on this set, CASF-2016, and a test set used for free energy perturbation (FEP) calculations, that not only highlights the competitive performance of AEV-PLIG, but provides a realistic assessment of machine learning models with rigorous physics-based approaches. Moreover, we demonstrate how leveraging augmented data (generated using template-based modelling or molecular docking) can significantly improve binding affinity prediction correlation and ranking on the FEP benchmark (weighted mean PCC and Kendall's τ increases from 0.41 and 0.26 to 0.59 and 0.42). These strategies together are closing the performance gap with FEP calculations (FEP+ achieves weighted mean PCC and Kendall's τ of 0.68 and 0.49 on the FEP benchmark) while being  ~400,000 times faster.

The algebraic extended atom-type graph-based model for precise ligand-receptor binding affinity prediction

Accurate prediction of ligand-receptor binding affinity is crucial in structure-based drug design, significantly impacting the development of effective drugs. Recent advances in machine learning (ML)-based scoring functions have improved these predictions, yet challenges remain in modeling complex molecular interactions. This study introduces the AGL-EAT-Score, a scoring function that integrates extended atom-type multiscale weighted colored subgraphs with algebraic graph theory. This approach leverages the eigenvalues and eigenvectors of graph Laplacian and adjacency matrices to capture high-level details of specific atom pairwise interactions. Evaluated against benchmark datasets such as CASF-2016, CASF-2013, and the Cathepsin S dataset, the AGL-EAT-Score demonstrates notable accuracy, outperforming existing traditional and ML-based methods. The model's strength lies in its comprehensive similarity analysis, examining protein sequence, ligand structure, and binding site similarities, thus ensuring minimal bias and over-representation in the training sets. The use of extended atom types in graph coloring enhances the model's capability to capture the intricacies of protein-ligand interactions. The AGL-EAT-Score marks a significant advancement in drug design, offering a tool that could potentially refine and accelerate the drug discovery process.Scientific Contribution The AGL-EAT-Score presents an algebraic graph-based framework that predicts ligand-receptor binding affinity by constructing multiscale weighted colored subgraphs from the 3D structure of protein-ligand complexes. It improves prediction accuracy by modeling interactions between extended atom types, addressing challenges like dataset bias and over-representation. Benchmark evaluations demonstrate that AGL-EAT-Score outperforms existing methods, offering a robust and systematic tool for structure-based drug design.

SG-ML-PLAP: A structure-guided machine learning-based scoring function for protein-ligand binding affinity prediction

Computational methods to predict binding affinity of protein-ligand complex have been used extensively to design inhibitors for proteins selected as drug targets. In recent years machine learning (ML) is being increasingly used for design of drugs/inhibitors. However, ranking compounds as per their experimental binding affinity has remained a major challenge. Therefore, it is necessary to develop ML-based scoring function (MLSF) for predicting the binding affinity of protein-ligand complexes. In this work, protein-ligand interaction features, namely, extended connectivity interaction fingerprints (ECIF), derived from the PDBbind dataset have been used to build ML models for binding affinity prediction. The benchmarking has been done on the Comparative Assessment of Scoring Functions (CASF) dataset and also by predicting the binding affinity of unseen protein-ligand complexes which have structural features different from those present in the training dataset. Furthermore, an improvement in the performance of MLSF on the redocked CASF complexes generated by AutoDock Vina software was seen when the training set consisting of crystal structures was supplemented with redocked protein-ligand complexes. The MLSF trained on crystal structures alone using a combination of ECIF and VINA features also predicted binding affinities of crystal as well as docked complexes with high accuracy. Overall, the MLSF developed in this work shows improved performance compared to conventional SFs and several other MLSFs. It will be a valuable resource for identifying novel inhibitors by structure-based virtual screening protocols. The proposed MLSF SG-ML-PLAP (Structure-Guided Machine-Learning-based Protein-Ligand Affinity Predictor) is freely accessible as a webserver, http://www.nii.ac.in/sg-ml-plap.html.

Enhancing protein-ligand binding affinity prediction through sequential fusion of graph and convolutional neural networks

Predicting protein-ligand binding affinity is a crucial and challenging task in structure-based drug discovery. With the accumulation of complex structures and binding affinity data, various machine-learning scoring functions, particularly those based on deep learning, have been developed for this task, exhibiting superiority over their traditional counterparts. A fusion model sequentially connecting a graph neural network (GNN) and a convolutional neural network (CNN) to predict protein-ligand binding affinity is proposed in this work. In this model, the intermediate outputs of the GNN layers, as supplementary descriptors of atomic chemical environments at different levels, are concatenated with the input features of CNN. The model demonstrates a noticeable improvement in performance on CASF-2016 benchmark compared to its constituent CNN models. The generalization ability of the model is evaluated by setting a series of thresholds for ligand extended-connectivity fingerprint similarity or protein sequence similarity between the training and test sets. Masking experiment reveals that model can capture key interaction regions. Furthermore, the fusion model is applied to a virtual screening task for a novel target, PI5P4Kα. The fusion strategy significantly improves the ability of the constituent CNN model to identify active compounds. This work offers a novel approach to enhancing the accuracy of deep learning models in predicting binding affinity through fusion strategies.

Interaction-Based Inductive Bias in Graph Neural Networks: Enhancing Protein-Ligand Binding Affinity Predictions From 3D Structures

Inductive bias in machine learning (ML) is the set of assumptions describing how a model makes predictions. Different ML-based methods for protein-ligand binding affinity (PLA) prediction have different inductive biases, leading to different levels of generalization capability and interpretability. Intuitively, the inductive bias of an ML-based model for PLA prediction should fit in with biological mechanisms relevant for binding to achieve good predictions with meaningful reasons. To this end, we propose an interaction-based inductive bias to restrict neural networks to functions relevant for binding with two assumptions: 1) A protein-ligand complex can be naturally expressed as a heterogeneous graph with covalent and non-covalent interactions; 2) The predicted PLA is the sum of pairwise atom-atom affinities determined by non-covalent interactions. The interaction-based inductive bias is embodied by an explainable heterogeneous interaction graph neural network (EHIGN) for explicitly modeling pairwise atom-atom interactions to predict PLA from 3D structures. Extensive experiments demonstrate that EHIGN achieves better generalization capability than other state-of-the-art ML-based baselines in PLA prediction and structure-based virtual screening. More importantly, comprehensive analyses of distance-affinity, pose-affinity, and substructure-affinity relations suggest that the interaction-based inductive bias can guide the model to learn atomic interactions that are consistent with physical reality. As a case study to demonstrate practical usefulness, our method is tested for predicting the efficacy of Nirmatrelvir against SARS-CoV-2 variants. EHIGN successfully recognizes the changes in the efficacy of Nirmatrelvir for different SARS-CoV-2 variants with meaningful reasons.

Structure-Aware Graph Attention Diffusion Network for Protein-Ligand Binding Affinity Prediction

Accurate prediction of protein-ligand binding affinities can significantly advance the development of drug discovery. Several graph neural network (GNN)-based methods learn representations of protein-ligand complexes via modeling intermolecule interactions and spatial structures (e.g., distances and angles) of complexes. However, these methods fail to emphasize the importance of bonds and learn hierarchical structures of complexes, which are significant for binding affinity prediction. In this article, we propose the structure-aware graph attention diffusion network (SGADN) to incorporate both distance and angle information for efficient spatial structure learning. We model complexes as line graphs with distance and angle information, focusing on bonds as nodes. Then we perform line graph attention diffusion layers (LGADLs) on line graphs to explore long-range bond node interactions and enhance spatial structure learning. Furthermore, we propose an attentive pooling layer (APL) to refine the hierarchical structures in complexes. Extensive experimental studies on two benchmarks demonstrate the superiority of SGADN for binding affinity prediction.

Exploring the potential of compound-protein complex structure-free models in virtual screening using BlendNet

Identifying new compounds that interact with a target is a crucial time-limiting step in the initial phases of drug discovery. Compound-protein complex structure-based affinity prediction models can expedite this process; however, their dependence on high-quality three-dimensional (3D) complex structures limits their practical application. Prediction models that do not require 3D complex structures for binding-affinity estimation offer a theoretically attractive alternative; however, accurately predicting affinity without interaction information presents significant challenges. We introduce BlendNet, a framework that employs a knowledge transfer strategy to improve affinity prediction accuracy by learning the interdependent relationships between compounds and proteins without relying on 3D complex structures. Compared with state-of-the-art models for affinity prediction, BlendNet demonstrated superior performance across various cold-start cases. The ability of BlendNet to interpret compound-protein interactions without utilizing complex structure data highlights its potential to accelerate and streamline drug development.

Predicting Protein-Ligand Binding Affinity Using Fusion Model of Spatial-Temporal Graph Neural Network and 3D Structure-Based Complex Graph

The investigation of molecular interactions between ligands and their target molecules is becoming more significant as protein structure data continues to develop. In this study, we introduce PLA-STGCNnet, a deep fusion spatial-temporal graph neural network designed to study protein-ligand interactions based on the 3D structural data of protein-ligand complexes. Unlike 1D protein sequences or 2D ligand graphs, the 3D graph representation offers a more precise portrayal of the complex interactions between proteins and ligands. Research studies have shown that our fusion model, PLA-STGCNnet, outperforms individual algorithms in accurately predicting binding affinity. The advantage of a fusion model is the ability to fully combine the advantages of multiple different models and improve overall performance by combining their features and outputs. Our fusion model shows satisfactory performance on different data sets, which proves its generalization ability and stability. The fusion-based model showed good performance in protein-ligand affinity prediction, and we successfully applied the model to drug screening. Our research underscores the promise of fusion spatial-temporal graph neural networks in addressing complex challenges in protein-ligand affinity prediction. The Python scripts for implementing various model components are accessible at https://github.com/ligaili01/PLA-STGCN.

BASE: a web service for providing compound-protein binding affinity prediction datasets with reduced similarity bias

BACKGROUND

Deep learning-based drug-target affinity (DTA) prediction methods have shown impressive performance, despite a high number of training parameters relative to the available data. Previous studies have highlighted the presence of dataset bias by suggesting that models trained solely on protein or ligand structures may perform similarly to those trained on complex structures. However, these studies did not propose solutions and focused solely on analyzing complex structure-based models. Even when ligands are excluded, protein-only models trained on complex structures still incorporate some ligand information at the binding sites. Therefore, it is unclear whether binding affinity can be accurately predicted using only compound or protein features due to potential dataset bias. In this study, we expanded our analysis to comprehensive databases and investigated dataset bias through compound and protein feature-based methods using multilayer perceptron models. We assessed the impact of this bias on current prediction models and proposed the binding affinity similarity explorer (BASE) web service, which provides bias-reduced datasets.

RESULTS

By analyzing eight binding affinity databases using multilayer perceptron models, we confirmed a bias where the compound-protein binding affinity can be accurately predicted using compound features alone. This bias arises because most compounds show consistent binding affinities due to high sequence or functional similarity among their target proteins. Our Uniform Manifold Approximation and Projection analysis based on compound fingerprints further revealed that low and high variation compounds do not exhibit significant structural differences. This suggests that the primary factor driving the consistent binding affinities is protein similarity rather than compound structure. We addressed this bias by creating datasets with progressively reduced protein similarity between the training and test sets, observing significant changes in model performance. We developed the BASE web service to allow researchers to download and utilize these datasets. Feature importance analysis revealed that previous models heavily relied on protein features. However, using bias-reduced datasets increased the importance of compound and interaction features, enabling a more balanced extraction of key features.

CONCLUSIONS

We propose the BASE web service, providing both the affinity prediction results of existing models and bias-reduced datasets. These resources contribute to the development of generalized and robust predictive models, enhancing the accuracy and reliability of DTA predictions in the drug discovery process. BASE is freely available online at https://synbi2024.kaist.ac.kr/base .

From Static to Dynamic Structures: Improving Binding Affinity Prediction with Graph-Based Deep Learning

Accurate prediction of protein-ligand binding affinities is an essential challenge in structure-based drug design. Despite recent advances in data-driven methods for affinity prediction, their accuracy is still limited, partially because they only take advantage of static crystal structures while the actual binding affinities are generally determined by the thermodynamic ensembles between proteins and ligands. One effective way to approximate such a thermodynamic ensemble is to use molecular dynamics (MD) simulation. Here, an MD dataset containing 3,218 different protein-ligand complexes is curated, and Dynaformer, a graph-based deep learning model is further developed to predict the binding affinities by learning the geometric characteristics of the protein-ligand interactions from the MD trajectories. In silico experiments demonstrated that the model exhibits state-of-the-art scoring and ranking power on the CASF-2016 benchmark dataset, outperforming the methods hitherto reported. Moreover, in a virtual screening on heat shock protein 90 (HSP90) using Dynaformer, 20 candidates are identified and their binding affinities are further experimentally validated. Dynaformer displayed promising results in virtual drug screening, revealing 12 hit compounds (two are in the submicromolar range), including several novel scaffolds. Overall, these results demonstrated that the approach offer a promising avenue for accelerating the early drug discovery process.

Proximity Graph Networks: Predicting Ligand Affinity with Message Passing Neural Networks

Message passing neural networks (MPNNs) on molecular graphs generate continuous and differentiable encodings of small molecules with state-of-the-art performance on protein-ligand complex scoring tasks. Here, we describe the proximity graph network (PGN) package, an open-source toolkit that constructs ligand-receptor graphs based on atom proximity and allows users to rapidly apply and evaluate MPNN architectures for a broad range of tasks. We demonstrate the utility of PGN by introducing benchmarks for affinity and docking score prediction tasks. Graph networks generalize better than fingerprint-based models and perform strongly for the docking score prediction task. Overall, MPNNs with proximity graph data structures augment the prediction of ligand-receptor complex properties when ligand-receptor data are available.

Target-specific novel molecules with their recipe: Incorporating synthesizability in the design process

Application of Artificial intelligence (AI) in drug discovery has led to several success stories in recent times. While traditional methods mostly relied upon screening large chemical libraries for early-stage drug-design, de novo design can help identify novel target-specific molecules by sampling from a much larger chemical space. Although this has increased the possibility of finding diverse and novel molecules from previously unexplored chemical space, this has also posed a great challenge for medicinal chemists to synthesize at least some of the de novo designed novel molecules for experimental validation. To address this challenge, in this work, we propose a novel forward synthesis-based generative AI method, which is used to explore the synthesizable chemical space. The method uses a structure-based drug design framework, where the target protein structure and a target-specific seed fragment from co-crystal structures can be the initial inputs. A random fragment from a purchasable fragment library can also be the input if a target-specific fragment is unavailable. Then a template-based forward synthesis route prediction and molecule generation is performed in parallel using the Monte Carlo Tree Search (MCTS) method where, the subsequent fragments for molecule growth can again be obtained from a purchasable fragment library. The rewards for each iteration of MCTS are computed using a drug-target affinity (DTA) model based on the docking pose of the generated reaction intermediates at the binding site of the target protein of interest. With the help of the proposed method, it is now possible to overcome one of the major obstacles posed to the AI-based drug design approaches through the ability of the method to design novel target-specific synthesizable molecules.

De novo drug design as GPT language modeling: large chemistry models with supervised and reinforcement learning

In recent years, generative machine learning algorithms have been successful in designing innovative drug-like molecules. SMILES is a sequence-like language used in most effective drug design models. Due to data's sequential structure, models such as recurrent neural networks and transformers can design pharmacological compounds with optimized efficacy. Large language models have advanced recently, but their implications on drug design have not yet been explored. Although one study successfully pre-trained a large chemistry model (LCM), its application to specific tasks in drug discovery is unknown. In this study, the drug design task is modeled as a causal language modeling problem. Thus, the procedure of reward modeling, supervised fine-tuning, and proximal policy optimization was used to transfer the LCM to drug design, similar to Open AI's ChatGPT and InstructGPT procedures. By combining the SMILES sequence with chemical descriptors, the novel efficacy evaluation model exceeded its performance compared to previous studies. After proximal policy optimization, the drug design model generated molecules with 99.2% having efficacy pIC50 > 7 towards the amyloid precursor protein, with 100% of the generated molecules being valid and novel. This demonstrated the applicability of LCMs in drug discovery, with benefits including less data consumption while fine-tuning. The applicability of LCMs to drug discovery opens the door for larger studies involving reinforcement-learning with human feedback, where chemists provide feedback to LCMs and generate higher-quality molecules. LCMs' ability to design similar molecules from datasets paves the way for more accessible, non-patented alternatives to drug molecules.

Water Network-Augmented Two-State Model for Protein-Ligand Binding Affinity Prediction

Water network rearrangement from the ligand-unbound state to the ligand-bound state is known to have significant effects on the protein-ligand binding interactions, but most of the current machine learning-based scoring functions overlook these effects. In this study, we endeavor to construct a comprehensive and realistic deep learning model by incorporating water network information into both ligand-unbound and -bound states. In particular, extended connectivity interaction features were integrated into graph representation, and graph transformer operator was employed to extract features of the ligand-unbound and -bound states. Through these efforts, we developed a water network-augmented two-state model called ECIFGraph::HM-Holo-Apo. Our new model exhibits satisfactory performance in terms of scoring, ranking, docking, screening, and reverse screening power tests on the CASF-2016 benchmark. In addition, it can achieve superior performance in large-scale docking-based virtual screening tests on the DEKOIS2.0 data set. Our study highlights that the use of a water network-augmented two-state model can be an effective strategy to bolster the robustness and applicability of machine learning-based scoring functions, particularly for targets with hydrophilic or solvent-exposed binding pockets.

An ensemble-based approach to estimate confidence of predicted protein-ligand binding affinity values

When designing a machine learning-based scoring function, we access a limited number of protein-ligand complexes with experimentally determined binding affinity values, representing only a fraction of all possible protein-ligand complexes. Consequently, it is crucial to report a measure of confidence and quantify the uncertainty in the model's predictions during test time. Here, we adopt the conformal prediction technique to evaluate the confidence of a prediction for each member of the core set of the CASF 2016 benchmark. The conformal prediction technique requires a diverse ensemble of predictors for uncertainty estimation. To this end, we introduce ENS-Score as an ensemble predictor, which includes 30 models with different protein-ligand representation approaches and achieves Pearson's correlation of 0.842 on the core set of the CASF 2016 benchmark. Also, we comprehensively investigate the residual error of each data point to assess the normality behavior of the distribution of the residual errors and their correlation to the structural features of the ligands, such as hydrophobic interactions and halogen bonding. In the end, we provide a local host web application to facilitate the usage of ENS-Score. All codes to repeat results are provided at https://github.com/miladrayka/ENS_Score.

Directional ΔG Neural Network (DrΔG-Net): A Modular Neural Network Approach to Binding Free Energy Prediction

The protein-ligand binding free energy is a central quantity in structure-based computational drug discovery efforts. Although popular alchemical methods provide sound statistical means of computing the binding free energy of a large breadth of systems, they are generally too costly to be applied at the same frequency as end point or ligand-based methods. By contrast, these data-driven approaches are typically fast enough to address thousands of systems but with reduced transferability to unseen systems. We introduce DrΔG-Net (or simply Dragnet), an equivariant graph neural network that can blend ligand-based and protein-ligand data-driven approaches. It is based on a 3D fingerprint representation of the ligand alone and in complex with the protein target. Dragnet is a global scoring function to predict the binding affinity of arbitrary protein-ligand complexes, but can be easily tuned via transfer learning to specific systems or end points, performing similarly to common 2D ligand-based approaches in these tasks. Dragnet is evaluated on a total of 28 validation proteins with a set of congeneric ligands derived from the Binding DB and one custom set extracted from the ChEMBL Database. In general, a handful of experimental binding affinities are sufficient to optimize the scoring function for a particular protein and ligand scaffold. When not available, predictions from physics-based methods such as absolute free energy perturbation can be used for the transfer learning tuning of Dragnet. Furthermore, we use our data to illustrate the present limitations of data-driven modeling of binding free energy predictions.

Machine Learning for Sequence and Structure-Based Protein-Ligand Interaction Prediction

Developing new drugs is too expensive and time -consuming. Accurately predicting the interaction between drugs and targets will likely change how the drug is discovered. Machine learning-based protein-ligand interaction prediction has demonstrated significant potential. In this paper, computational methods, focusing on sequence and structure to study protein-ligand interactions, are examined. Therefore, this paper starts by presenting an overview of the data sets applied in this area, as well as the various approaches applied for representing proteins and ligands. Then, sequence-based and structure-based classification criteria are subsequently utilized to categorize and summarize both the classical machine learning models and deep learning models employed in protein-ligand interaction studies. Moreover, the evaluation methods and interpretability of these models are proposed. Furthermore, delving into the diverse applications of protein-ligand interaction models in drug research is presented. Lastly, the current challenges and future directions in this field are addressed.

Drugging the entire human proteome: Are we there yet?

Each of the ∼20,000 proteins in the human proteome is a potential target for compounds that bind to it and modify its function. The 3D structures of most of these proteins are now available. Here, we discuss the prospects for using these structures to perform proteome-wide virtual HTS (VHTS). We compare physics-based (docking) and AI VHTS approaches, some of which are now being applied with large databases of compounds to thousands of targets. Although preliminary proteome-wide screens are now within our grasp, further methodological developments are expected to improve the accuracy of the results.

Improving structure-based protein-ligand affinity prediction by graph representation learning and ensemble learning

Predicting protein-ligand binding affinity presents a viable solution for accelerating the discovery of new lead compounds. The recent widespread application of machine learning approaches, especially graph neural networks, has brought new advancements in this field. However, some existing structure-based methods treat protein macromolecules and ligand small molecules in the same way and ignore the data heterogeneity, potentially leading to incomplete exploration of the biochemical information of ligands. In this work, we propose LGN, a graph neural network-based fusion model with extra ligand feature extraction to effectively capture local features and global features within the protein-ligand complex, and make use of interaction fingerprints. By combining the ligand-based features and interaction fingerprints, LGN achieves Pearson correlation coefficients of up to 0.842 on the PDBbind 2016 core set, compared to 0.807 when using the features of complex graphs alone. Finally, we verify the rationalization and generalization of our model through comprehensive experiments. We also compare our model with state-of-the-art baseline methods, which validates the superiority of our model. To reduce the impact of data similarity, we increase the robustness of the model by incorporating ensemble learning.

Structure-based, deep-learning models for protein-ligand binding affinity prediction

The launch of AlphaFold series has brought deep-learning techniques into the molecular structural science. As another crucial problem, structure-based prediction of protein-ligand binding affinity urgently calls for advanced computational techniques. Is deep learning ready to decode this problem? Here we review mainstream structure-based, deep-learning approaches for this problem, focusing on molecular representations, learning architectures and model interpretability. A model taxonomy has been generated. To compensate for the lack of valid comparisons among those models, we realized and evaluated representatives from a uniform basis, with the advantages and shortcomings discussed. This review will potentially benefit structure-based drug discovery and related areas.

Decision tree-based identification of important molecular fragments for protein-ligand binding

Fragment-based drug design is an emerging technology in pharmaceutical research and development. One of the key aspects of this technology is the identification and quantitative characterization of molecular fragments. This study presents a strategy for identifying important molecular fragments based on molecular fingerprints and decision tree algorithms and verifies its feasibility in predicting protein-ligand binding affinity. Specifically, the three-dimensional (3D) structures of protein-ligand complexes are encoded using extended-connectivity fingerprints (ECFP), and three decision tree models, namely Random Forest, XGBoost, and LightGBM, are used to quantitatively characterize the feature importance, thereby extracting important molecular fragments with high reliability. Few-shot learning reveals that the extracted molecular fragments contribute significantly and consistently to the binding affinity even with a small sample size. Despite the absence of location and distance information for molecular fragments in ECFP, 3D visualization, in combination with the reverse ECFP process, shows that the majority of the extracted fragments are located at the binding interface of the protein and the ligand. This alignment with the distance constraints critical for binding affinity further supports the reliability of the strategy for identifying important molecular fragments.

Ensemble of local and global information for Protein-Ligand Binding Affinity Prediction

Accurately predicting protein-ligand binding affinities is crucial for determining molecular properties and understanding their physical effects. Neural networks and transformers are the predominant methods for sequence modeling, and both have been successfully applied independently for protein-ligand binding affinity prediction. As local and global information of molecules are vital for protein-ligand binding affinity prediction, we aim to combine bi-directional gated recurrent unit (BiGRU) and convolutional neural network (CNN) to effectively capture both local and global molecular information. Additionally, attention mechanisms can be incorporated to automatically learn and adjust the level of attention given to local and global information, thereby enhancing the performance of the model. To achieve this, we propose the PLAsformer approach, which encodes local and global information of molecules using 3DCNN and BiGRU with attention mechanism, respectively. This approach enhances the model's ability to encode comprehensive local and global molecular information. PLAsformer achieved a Pearson's correlation coefficient of 0.812 and a Root Mean Square Error (RMSE) of 1.284 when comparing experimental and predicted affinity on the PDBBind-2016 dataset. These results surpass the current state-of-the-art methods for binding affinity prediction. The high accuracy of PLAsformer's predictive performance, along with its excellent generalization ability, is clearly demonstrated by these findings.

A practical guide to machine-learning scoring for structure-based virtual screening

Structure-based virtual screening (SBVS) via docking has been used to discover active molecules for a range of therapeutic targets. Chemical and protein data sets that contain integrated bioactivity information have increased both in number and in size. Artificial intelligence and, more concretely, its machine-learning (ML) branch, including deep learning, have effectively exploited these data sets to build scoring functions (SFs) for SBVS against targets with an atomic-resolution 3D model (e.g., generated by X-ray crystallography or predicted by AlphaFold2). Often outperforming their generic and non-ML counterparts, target-specific ML-based SFs represent the state of the art for SBVS. Here, we present a comprehensive and user-friendly protocol to build and rigorously evaluate these new SFs for SBVS. This protocol is organized into four sections: (i) using a public benchmark of a given target to evaluate an existing generic SF; (ii) preparing experimental data for a target from public repositories; (iii) partitioning data into a training set and a test set for subsequent target-specific ML modeling; and (iv) generating and evaluating target-specific ML SFs by using the prepared training-test partitions. All necessary code and input/output data related to three example targets (acetylcholinesterase, HMG-CoA reductase, and peroxisome proliferator-activated receptor-α) are available at https://github.com/vktrannguyen/MLSF-protocol , can be run by using a single computer within 1 week and make use of easily accessible software/programs (e.g., Smina, CNN-Score, RF-Score-VS and DeepCoy) and web resources. Our aim is to provide practical guidance on how to augment training data to enhance SBVS performance, how to identify the most suitable supervised learning algorithm for a data set, and how to build an SF with the highest likelihood of discovering target-active molecules within a given compound library.

Multi-shelled ECIF: improved extended connectivity interaction features for accurate binding affinity prediction

Motivation

Extended connectivity interaction features (ECIF) is a method developed to predict protein-ligand binding affinity, allowing for detailed atomic representation. It performed very well in terms of Comparative Assessment of Scoring Functions 2016 (CASF-2016) scoring power. However, ECIF has the limitation of not being able to adequately account for interatomic distances.

Results

To investigate what kind of distance representation is effective for P-L binding affinity prediction, we have developed two algorithms that improved ECIF's feature extraction method to take distance into account. One is multi-shelled ECIF, which takes into account the distance between atoms by dividing the distance between atoms into multiple layers. The other is weighted ECIF, which weights the importance of interactions according to the distance between atoms. A comparison of these two methods shows that multi-shelled ECIF outperforms weighted ECIF and the original ECIF, achieving a CASF-2016 scoring power Pearson correlation coefficient of 0.877.

Availability and implementation

All the codes and data are available on GitHub (https://github.com/koji11235/MSECIFv2).

Geometric graph learning with extended atom-types features for protein-ligand binding affinity prediction

Understanding and accurately predicting protein-ligand binding affinity are essential in the drug design and discovery process. At present, machine learning-based methodologies are gaining popularity as a means of predicting binding affinity due to their efficiency and accuracy, as well as the increasing availability of structural and binding affinity data for protein-ligand complexes. In biomolecular studies, graph theory has been widely applied since graphs can be used to model molecules or molecular complexes in a natural manner. In the present work, we upgrade the graph-based learners for the study of protein-ligand interactions by integrating extensive atom types such as SYBYL and extended connectivity interactive features (ECIF) into multiscale weighted colored graphs (MWCG). By pairing with the gradient boosting decision tree (GBDT) machine learning algorithm, our approach results in two different methods, namely sybylGGL-Score and ecifGGL-Score. Both of our models are extensively validated in their scoring power using three commonly used benchmark datasets in the drug design area, namely CASF-2007, CASF-2013, and CASF-2016. The performance of our best model sybylGGL-Score is compared with other state-of-the-art models in the binding affinity prediction for each benchmark. While both of our models achieve state-of-the-art results, the SYBYL atom-type model sybylGGL-Score outperforms other methods by a wide margin in all benchmarks. Finally, the best-performing SYBYL atom-type model is evaluated on two test sets that are independent of CASF benchmarks.

Protein-ligand binding affinity prediction exploiting sequence constituent homology

MOTIVATION

Molecular docking is a commonly used approach for estimating binding conformations and their resultant binding affinities. Machine learning has been successfully deployed to enhance such affinity estimations. Many methods of varying complexity have been developed making use of some or all the spatial and categorical information available in these structures. The evaluation of such methods has mainly been carried out using datasets from PDBbind. Particularly the Comparative Assessment of Scoring Functions (CASF) 2007, 2013, and 2016 datasets with dedicated test sets. This work demonstrates that only a small number of simple descriptors is necessary to efficiently estimate binding affinity for these complexes without the need to know the exact binding conformation of a ligand.

RESULTS

The developed approach of using a small number of ligand and protein descriptors in conjunction with gradient boosting trees demonstrates high performance on the CASF datasets. This includes the commonly used benchmark CASF2016 where it appears to perform better than any other approach. This methodology is also useful for datasets where the spatial relationship between the ligand and protein is unknown as demonstrated using a large ChEMBL-derived dataset.

AVAILABILITY AND IMPLEMENTATION

Code and data uploaded to https://github.com/abbiAR/PLBAffinity.

AQDnet: Deep Neural Network for Protein-Ligand Docking Simulation

We have developed an innovative system, AI QM Docking Net (AQDnet), which utilizes the three-dimensional structure of protein-ligand complexes to predict binding affinity. This system is novel in two respects: first, it significantly expands the training dataset by generating thousands of diverse ligand configurations for each protein-ligand complex and subsequently determining the binding energy of each configuration through quantum computation. Second, we have devised a method that incorporates the atom-centered symmetry function (ACSF), highly effective in describing molecular energies, for the prediction of protein-ligand interactions. These advancements have enabled us to effectively train a neural network to learn the protein-ligand quantum energy landscape (P-L QEL). Consequently, we have achieved a 92.6% top 1 success rate in the CASF-2016 docking power, placing first among all models assessed in the CASF-2016, thus demonstrating the exceptional docking performance of our model.

SS-GNN: A Simple-Structured Graph Neural Network for Affinity Prediction

Efficient and effective drug-target binding affinity (DTBA) prediction is a challenging task due to the limited computational resources in practical applications and is a crucial basis for drug screening. Inspired by the good representation ability of graph neural networks (GNNs), we propose a simple-structured GNN model named SS-GNN to accurately predict DTBA. By constructing a single undirected graph based on a distance threshold to represent protein-ligand interactions, the scale of the graph data is greatly reduced. Moreover, ignoring covalent bonds in the protein further reduces the computational cost of the model. The graph neural network-multilayer perceptron (GNN-MLP) module takes the latent feature extraction of atoms and edges in the graph as two mutually independent processes. We also develop an edge-based atom-pair feature aggregation method to represent complex interactions and a graph pooling-based method to predict the binding affinity of the complex. We achieve state-of-the-art prediction performance using a simple model (with only 0.6 M parameters) without introducing complicated geometric feature descriptions. SS-GNN achieves Pearson's R_p = 0.853 on the PDBbind v2016 core set, outperforming state-of-the-art GNN-based methods by 5.2%. Moreover, the simplified model structure and concise data processing procedure improve the prediction efficiency of the model. For a typical protein-ligand complex, affinity prediction takes only 0.2 ms. All codes are freely accessible at https://github.com/xianyuco/SS-GNN.

A Simple Way to Incorporate Target Structural Information in Molecular Generative Models

Deep learning generative models are now being applied in various fields including drug discovery. In this work, we propose a novel approach to include target 3D structural information in molecular generative models for structure-based drug design. The method combines a message-passing neural network model that predicts docking scores with a generative neural network model as its reward function to navigate the chemical space searching for molecules that bind favorably with a specific target. A key feature of the method is the construction of target-specific molecular sets for training, designed to overcome potential transferability issues of surrogate docking models through a two-round training process. Consequently, this enables accurate guided exploration of the chemical space without reliance on the collection of prior knowledge about active and inactive compounds for the specific target. Tests on eight target proteins showed a 100-fold increase in hit generation compared to conventional docking calculations and the ability to generate molecules similar to approved drugs or known active ligands for specific targets without prior knowledge. This method provides a general and highly efficient solution for structure-based molecular generation.

Geometric Interaction Graph Neural Network for Predicting Protein-Ligand Binding Affinities from 3D Structures (GIGN)

Predicting protein-ligand binding affinities (PLAs) is a core problem in drug discovery. Recent advances have shown great potential in applying machine learning (ML) for PLA prediction. However, most of them omit the 3D structures of complexes and physical interactions between proteins and ligands, which are considered essential to understanding the binding mechanism. This paper proposes a geometric interaction graph neural network (GIGN) that incorporates 3D structures and physical interactions for predicting protein-ligand binding affinities. Specifically, we design a heterogeneous interaction layer that unifies covalent and noncovalent interactions into the message passing phase to learn node representations more effectively. The heterogeneous interaction layer also follows fundamental biological laws, including invariance to translations and rotations of the complexes, thus avoiding expensive data augmentation strategies. GIGN achieves state-of-the-art performance on three external test sets. Moreover, by visualizing learned representations of protein-ligand complexes, we show that the predictions of GIGN are biologically meaningful.

GB-score: Minimally designed machine learning scoring function based on distance-weighted interatomic contact features

In recent years, thanks to advances in computer hardware and dataset availability, data-driven approaches (like machine learning) have become one of the essential parts of the drug design framework to accelerate drug discovery procedures. Constructing a new scoring function, a function that can predict the binding score for a generated protein-ligand pose during docking procedure or a crystal complex, based on machine and deep learning has become an active research area in computer-aided drug design. GB-Score is a state-of-the-art machine learning-based scoring function that utilizes distance-weighted interatomic contact features, PDBbind-v2019 general set, and Gradient Boosting Trees algorithm to the binding affinity prediction. The distance-weighted interatomic contact featurization method used the distance between different ligand and protein atom types for numerical representation of the protein-ligand complex. GB-Score attains Pearson's correlation 0.862 and RMSE 1.190 on the CASF-2016 benchmark test in the scoring power metric. GB-Score's codes are freely available on the web at https://github.com/miladrayka/GB_Score.

Consensus Virtual Screening Protocol Towards the Identification of Small Molecules Interacting with the Colchicine Binding Site of the Tubulin-microtubule System

Modification of the tubulin-microtubule (Tub-Mts) system has generated effective strategies for developing different treatments for cancer. A huge amount of clinical data about inhibitors of the tubulin-microtubule system have supported and validated the studies on this pharmacological target. However, many tubulin-microtubule inhibitors have been developed from representative and common scaffolds that cover a small region of the chemical space with limited structural innovation. The main goal of this study is to develop the first consensus virtual screening protocol for natural products (ligand- and structure-based drug design methods) tuned for the identification of new potential inhibitors of the Tub-Mts system. A combined strategy that involves molecular similarity, molecular docking, pharmacophore modeling, and in silico ADMET prediction has been employed to prioritize the selections of potential inhibitors of the Tub-Mts system. Five compounds were selected and further studied using molecular dynamics and binding energy predictions to characterize their possible binding mechanisms. Their structures correspond to 5-[2-(4-hydroxy-3-methoxyphenyl) ethyl]-2,3-dimethoxyphenol (1), 9,10-dihydro-3,4-dimethoxy-2,7-phenanthrenediol (2), 2-(3,4-dimethoxyphenyl)-5,7-dihydroxy-6-methoxy-4H-1-benzopyran-4-one (3), 13,14-epoxyparvifoline-4',5',6'-trimethoxybenzoate (4), and phenylmethyl 6-hydroxy-2,3-dimethoxybenzoate (5). Compounds 1-3 have been associated with literature reports that confirm their activity against several cancer cell lines, thus supporting the utility of this protocol.

De novo design of anti-tuberculosis agents using a structure-based deep learning method

Mycobacterium tuberculosis (Mtb) is a pathogen of major concern due to its ability to withstand both first- and second-line antibiotics, leading to drug resistance. Thus, there is a critical need for identification of novel anti-tuberculosis agents targeting Mtb-specific proteins. The ceaseless search for novel antimicrobial agents to combat drug-resistant bacteria can be accelerated by the development of advanced deep learning methods, to explore both existing and uncharted regions of the chemical space. The adaptation of deep learning methods to under-explored pathogens such as Mtb is a challenging aspect, as most of the existing methods rely on the availability of sufficient target-specific ligand data to design novel small molecules with optimized bioactivity. In this work, we report the design of novel anti-tuberculosis agents targeting the Mtb chorismate mutase protein using a structure-based drug design algorithm. The structure-based deep learning method relies on the knowledge of the target protein's binding site structure alone for conditional generation of novel small molecules. The method eliminates the need for curation of a high-quality target-specific small molecule dataset, which remains a challenge even for many druggable targets, including Mtb chorismate mutase. Novel molecules are proposed, that show high complementarity to the target binding site. The graph attention model could identify the probable key binding site residues, which influenced the conditional molecule generator to design new molecules with pharmacophoric features similar to the known inhibitors.

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/ .

Learning from Docked Ligands: Ligand-Based Features Rescue Structure-Based Scoring Functions When Trained on Docked Poses

Machine learning scoring functions for protein-ligand binding affinity have been found to consistently outperform classical scoring functions when trained and tested on crystal structures of bound protein-ligand complexes. However, it is less clear how these methods perform when applied to docked poses of complexes. We explore how the use of docked rather than crystallographic poses for both training and testing affects the performance of machine learning scoring functions. Using the PDBbind Core Sets as benchmarks, we show that the performance of a structure-based machine learning scoring function trained and tested on docked poses is lower than that of the same scoring function trained and tested on crystallographic poses. We construct a hybrid scoring function by combining both structure-based and ligand-based features, and show that its ability to predict binding affinity using docked poses is comparable to that of purely structure-based scoring functions trained and tested on crystal poses. We also present a new, freely available validation set─the Updated DUD-E Diverse Subset─for binding affinity prediction using data from DUD-E and ChEMBL. Despite strong performance on docked poses of the PDBbind Core Sets, we find that our hybrid scoring function sometimes generalizes poorly to a protein target not represented in the training set, demonstrating the need for improved scoring functions and additional validation benchmarks.

Systematic Improvement of the Performance of Machine Learning Scoring Functions by Incorporating Features of Protein-Bound Water Molecules

Water molecules at the ligand-protein interfaces play crucial roles in the binding of the ligands, but the behavior of protein-bound water is largely ignored in many currently used machine learning (ML)-based scoring functions (SFs). In an attempt to improve the prediction performance of existing ML-based SFs, we estimated the water distribution with a HydraMap (HM) method and then incorporated the features extracted from protein-bound waters obtained in this way into three ML-based SFs: RF-Score, ECIF, and PLEC. It was found that a combination of HM-based features can consistently improve the performance of all three SFs, including their scoring, ranking, and docking power. HydraMap-based features show consistently good performance with both crystal structures and docked structures, demonstrating their robustness for SFs. Overall, HM-based features, which are a statistical representation of hydration sites at protein-ligand interfaces, are expected to improve the prediction performance for diverse SFs.

Progress and Impact of Latin American Natural Product Databases

Natural products (NPs) are a rich source of structurally novel molecules, and the chemical space they encompass is far from being fully explored. Over history, NPs have represented a significant source of bioactive molecules and have served as a source of inspiration for developing many drugs on the market. On the other hand, computer-aided drug design (CADD) has contributed to drug discovery research, mitigating costs and time. In this sense, compound databases represent a fundamental element of CADD. This work reviews the progress toward developing compound databases of natural origin, and it surveys computational methods, emphasizing chemoinformatic approaches to profile natural product databases. Furthermore, it reviews the present state of the art in developing Latin American NP databases and their practical applications to the drug discovery area.

SCORCH: Improving structure-based virtual screening with machine learning classifiers, data augmentation, and uncertainty estimation

INTRODUCTION

The discovery of a new drug is a costly and lengthy endeavour. The computational prediction of which small molecules can bind to a protein target can accelerate this process if the predictions are fast and accurate enough. Recent machine-learning scoring functions re-evaluate the output of molecular docking to achieve more accurate predictions. However, previous scoring functions were trained on crystalised protein-ligand complexes and datasets of decoys. The limited availability of crystal structures and biases in the decoy datasets can lower the performance of scoring functions.

OBJECTIVES

To address key limitations of previous scoring functions and thus improve the predictive performance of structure-based virtual screening.

METHODS

A novel machine-learning scoring function was created, named SCORCH (Scoring COnsensus for RMSD-based Classification of Hits). To develop SCORCH, training data is augmented by considering multiple ligand poses and labelling poses based on their RMSD from the native pose. Decoy bias is addressed by generating property-matched decoys for each ligand and using the same methodology for preparing and docking decoys and ligands. A consensus of 3 different machine learning approaches is also used to improve performance.

RESULTS

We find that multi-pose augmentation in SCORCH improves its docking power and screening power on independent benchmark datasets. SCORCH outperforms an equivalent scoring function trained on single poses, with a 1% enrichment factor (EF) of 13.78 vs. 10.86 on 18 DEKOIS 2.0 targets and a mean native pose rank of 5.9 vs 30.4 on CSAR 2014. Additionally, SCORCH outperforms widely used scoring functions in virtual screening and pose prediction on independent benchmark datasets.

CONCLUSION

By rationally addressing key limitations of previous scoring functions, SCORCH improves the performance of virtual screening. SCORCH also provides an estimate of its uncertainty, which can help reduce the cost and time required for drug discovery.

Protein-Ligand Docking in the Machine-Learning Era

Molecular docking plays a significant role in early-stage drug discovery, from structure-based virtual screening (VS) to hit-to-lead optimization, and its capability and predictive power is critically dependent on the protein-ligand scoring function. In this review, we give a broad overview of recent scoring function development, as well as the docking-based applications in drug discovery. We outline the strategies and resources available for structure-based VS and discuss the assessment and development of classical and machine learning protein-ligand scoring functions. In particular, we highlight the recent progress of machine learning scoring function ranging from descriptor-based models to deep learning approaches. We also discuss the general workflow and docking protocols of structure-based VS, such as structure preparation, binding site detection, docking strategies, and post-docking filter/re-scoring, as well as a case study on the large-scale docking-based VS test on the LIT-PCBA data set.

Predicting Protein-Ligand Docking Structure with Graph Neural Network

Modern day drug discovery is extremely expensive and time consuming. Although computational approaches help accelerate and decrease the cost of drug discovery, existing computational software packages for docking-based drug discovery suffer from both low accuracy and high latency. A few recent machine learning-based approaches have been proposed for virtual screening by improving the ability to evaluate protein-ligand binding affinity, but such methods rely heavily on conventional docking software to sample docking poses, which results in excessive execution latencies. Here, we propose and evaluate a novel graph neural network (GNN)-based framework, MedusaGraph, which includes both pose-prediction (sampling) and pose-selection (scoring) models. Unlike the previous machine learning-centric studies, MedusaGraph generates the docking poses directly and achieves from 10 to 100 times speedup compared to state-of-the-art approaches, while having a slightly better docking accuracy.

Structure-based virtual screening for PDL1 dimerizers: Evaluating generic scoring functions

The interaction between PD1 and its ligand PDL1 has been shown to render tumor cells resistant to apoptosis and promote tumor progression. An innovative mechanism to inhibit the PD1/PDL1 interaction is PDL1 dimerization induced by small-molecule PDL1 binders. Structure-based virtual screening is a promising approach to discovering such small-molecule PD1/PDL1 inhibitors. Here we investigate which type of generic scoring functions is most suitable to tackle this problem. We consider CNN-Score, an ensemble of convolutional neural networks, as the representative of machine-learning scoring functions. We also evaluate Smina, a commonly used classical scoring function, and IFP, a top structural fingerprint similarity scoring function. These three types of scoring functions were evaluated on two test sets sharing the same set of small-molecule PD1/PDL1 inhibitors, but using different types of inactives: either true inactives (molecules with no in vitro PD1/PDL1 inhibition activity) or assumed inactives (property-matched decoy molecules generated from each active). On both test sets, CNN-Score performed much better than Smina, which in turn strongly outperformed IFP. The fact that the latter was the case, despite precluding any possibility of exploiting decoy bias, demonstrates the predictive value of CNN-Score for PDL1. These results suggest that re-scoring Smina-docked molecules with CNN-Score is a promising structure-based virtual screening method to discover new small-molecule inhibitors of this therapeutic target.

Machine-Learning- and Knowledge-Based Scoring Functions Incorporating Ligand and Protein Fingerprints

We propose a novel machine-learning-based scoring function for drug discovery that incorporates ligand and protein structural information into a knowledge-based PMF score. Molecular docking, a simulation method for structure-based drug design (SBDD), is expected to reduce the enormous costs associated with conventional experimental methods in terms of rational drug discovery. Molecular docking has two main purposes: to predict ligand-binding structures for target proteins and to predict protein-ligand binding affinity. Currently available programs of molecular docking offer an accurate prediction of ligand binding structures for many systems. However, the accurate prediction of binding affinity remains challenging. In this study, we developed a new scoring function that incorporates fingerprints representing ligand and protein structures as descriptors in the PMF score. Here, regression analysis of the scoring function was performed using the following machine learning techniques: least absolute shrinkage and selection operator (LASSO) and light gradient boosting machine (LightGBM). The results on a test data set showed that the binding affinity delivered by the newly developed scoring function has a Pearson correlation coefficient of 0.79 with the experimental value, which surpasses that of the conventional scoring functions. Further analysis provided a chemical understanding of the descriptors that contributed significantly to the improvement in prediction accuracy. Our approach and findings are useful for rational drug discovery.

Delta Machine Learning to Improve Scoring-Ranking-Screening Performances of Protein-Ligand Scoring Functions

Protein-ligand scoring functions are widely used in structure-based drug design for fast evaluation of protein-ligand interactions, and it is of strong interest to develop scoring functions with machine-learning approaches. In this work, by expanding the training set, developing physically meaningful features, employing our recently developed linear empirical scoring function Lin_F9 (Yang, C. J. Chem. Inf. Model. 2021, 61, 4630-4644) as the baseline, and applying extreme gradient boosting (XGBoost) with Δ-machine learning, we have further improved the robustness and applicability of machine-learning scoring functions. Besides the top performances for scoring-ranking-screening power tests of the CASF-2016 benchmark, the new scoring function Δ_{Lin_F9}XGB also achieves superior scoring and ranking performances in different structure types that mimic real docking applications. The scoring powers of Δ_{Lin_F9}XGB for locally optimized poses, flexible redocked poses, and ensemble docked poses of the CASF-2016 core set achieve Pearson's correlation coefficient (R) values of 0.853, 0.839, and 0.813, respectively. In addition, the large-scale docking-based virtual screening test on the LIT-PCBA data set demonstrates the reliability and robustness of Δ_{Lin_F9}XGB in virtual screening application. The Δ_{Lin_F9}XGB scoring function and its code are freely available on the web at (https://yzhang.hpc.nyu.edu/Delta_LinF9_XGB).

A simple spatial extension to the extended connectivity interaction features for binding affinity prediction

The representation of the protein-ligand complexes used in building machine learning models play an important role in the accuracy of binding affinity prediction. The Extended Connectivity Interaction Features (ECIF) is one such representation. We report that (i) including the discretized distances between protein-ligand atom pairs in the ECIF scheme improves predictive accuracy, and (ii) in an evaluation using gradient boosted trees, we found that the resampling method used in selecting the best hyperparameters has a strong effect on predictive performance, especially for benchmarking purposes.

Dowker complex based machine learning (DCML) models for protein-ligand binding affinity prediction

With the great advancements in experimental data, computational power and learning algorithms, artificial intelligence (AI) based drug design has begun to gain momentum recently. AI-based drug design has great promise to revolutionize pharmaceutical industries by significantly reducing the time and cost in drug discovery processes. However, a major issue remains for all AI-based learning model that is efficient molecular representations. Here we propose Dowker complex (DC) based molecular interaction representations and Riemann Zeta function based molecular featurization, for the first time. Molecular interactions between proteins and ligands (or others) are modeled as Dowker complexes. A multiscale representation is generated by using a filtration process, during which a series of DCs are generated at different scales. Combinatorial (Hodge) Laplacian matrices are constructed from these DCs, and the Riemann zeta functions from their spectral information can be used as molecular descriptors. To validate our models, we consider protein-ligand binding affinity prediction. Our DC-based machine learning (DCML) models, in particular, DC-based gradient boosting tree (DC-GBT), are tested on three most-commonly used datasets, i.e., including PDBbind-2007, PDBbind-2013 and PDBbind-2016, and extensively compared with other existing state-of-the-art models. It has been found that our DC-based descriptors can achieve the state-of-the-art results and have better performance than all machine learning models with traditional molecular descriptors. Our Dowker complex based machine learning models can be used in other tasks in AI-based drug design and molecular data analysis.

With the ever-increasing accumulation of chemical and biomolecular data, data-driven artificial intelligence (AI) models will usher in an era of faster, cheaper and more-efficient drug design and drug discovery. However, unlike image, text, video, audio data, molecular data from chemistry and biology, have much complicated three-dimensional structures, as well as physical and chemical properties. Efficient molecular representations and descriptors are key to the success of machine learning models in drug design. Here, we propose Dowker complex based molecular representation and Riemann Zeta function based molecular featurization, for the first time. To characterize the complicated molecular structures and interactions at the atomic level, Dowker complexes are constructed. Based on them, intrinsic mathematical invariants are derived and used as molecular descriptors, which can be further combined with machine learning and deep learning models. Our model has achieved state-of-the-art results in protein-ligand binding affinity prediction, demonstrating its great potential for other drug design and discovery problems.

Affinity prediction using deep learning based on SMILES input for D3R grand challenge 4

Modern molecular docking comprises the prediction of pose and affinity. Prediction of docking poses is required for affinity prediction when three-dimensional coordinates of the ligand have not been provided. However, a large number of feature engineering is required for existing methods. In addition, there is a need for a robust model for the sequential combination of pose and affinity prediction due to the probabilistic deviation of the ligand position issue. We propose a pipeline using a bipartite graph neural network and transfer learning trained on a re-docking dataset. We evaluated our model on the released data from drug design data resource grand challenge 4 (D3R GC4). The two target protein data provided by the challenge have different patterns. The model outperformed the best participant by 9% on the BACE target protein from stage 2. Further, our model showed competitive performance on the CatS target protein.

Molecular persistent spectral image (Mol-PSI) representation for machine learning models in drug design

Artificial intelligence (AI)-based drug design has great promise to fundamentally change the landscape of the pharmaceutical industry. Even though there are great progress from handcrafted feature-based machine learning models, 3D convolutional neural networks (CNNs) and graph neural networks, effective and efficient representations that characterize the structural, physical, chemical and biological properties of molecular structures and interactions remain to be a great challenge. Here, we propose an equal-sized molecular 2D image representation, known as the molecular persistent spectral image (Mol-PSI), and combine it with CNN model for AI-based drug design. Mol-PSI provides a unique one-to-one image representation for molecular structures and interactions. In general, deep models are empowered to achieve better performance with systematically organized representations in image format. A well-designed parallel CNN architecture for adapting Mol-PSIs is developed for protein-ligand binding affinity prediction. Our results, for the three most commonly used databases, including PDBbind-v2007, PDBbind-v2013 and PDBbind-v2016, are better than all traditional machine learning models, as far as we know. Our Mol-PSI model provides a powerful molecular representation that can be widely used in AI-based drug design and molecular data analysis.

Protein-ligand binding affinity prediction based on profiles of intermolecular contacts

As a key element in structure-based drug design, binding affinity prediction (BAP) for putative protein-ligand complexes can be efficiently achieved by the incorporation of structural descriptors and machine-learning models. However, developing concise descriptors that will lead to accurate and interpretable BAP remains a difficult problem in this field. Herein, we introduce the profiles of intermolecular contacts (IMCPs) as descriptors for machine-learning-based BAP. IMCPs describe each group of protein-ligand contacts by the count and average distance of the group members, and collaborate closely with classical machine-learning models. Performed on multiple validation sets, IMCP-based models often result in better BAP accuracy than those originating from other similar descriptors. Additionally, IMCPs are simple and concise, and easy to interpret in model training. These descriptors highly conclude the structural information of protein-ligand complexes and can be easily updated with personalized profile features. IMCPs have been implemented in the BAP Toolkit on github ( https://github.com/debbydanwang/BAP).