A comprehensive survey on protein-ligand binding site prediction

Current Opinion in Structural Biology, Journal Year: 2024, Volume and Issue: 86, P. 102793 - 102793

Published: March 5, 2024

Language: Английский

---

AI-Driven Parametrization of Michaelis-Menten Maximal Velocity: Advancing in silico New Approach Methodologies (NAMs)

NAM journal., Journal Year: 2025, Volume and Issue: unknown, P. 100012 - 100012

Published: Feb. 1, 2025

Language: Английский

---

Ligandability and druggability assessment via machine learning

Wiley Interdisciplinary Reviews Computational Molecular Science, Journal Year: 2023, Volume and Issue: 13(5)

Published: June 4, 2023

Abstract Drug discovery is a daunting and failure‐prone task. A critical process in this research field represented by the biological target pocket identification steps as they heavily determine subsequent efforts selecting putative ligand, most often small molecule. Finding “ligandable” pockets, namely protein cavities that may accept drug‐like binder instrumental to more general drug oriented “druggability” estimation process. While high‐throughput experimental techniques exist identify binding sites other than orthosteric one, these are relatively expensive not so commonly available labs. In regard, computational means of detecting ligandable pockets advisable for their inexpensiveness speed. These methods can become, principle, particularly predictive when supported machine learning methodologies provide modeling framework. As with any data‐driven effort, outcome critically depends on input data, its featurization possible associated biases. Also, task, (supervised/unsupervised) method, usage molecular dynamics data considerably shape inherent assumptions step. Defining proper quantitative thermodynamic and/or kinetic score (or label) key process; here we revise literature propose residence time novel ideal indicator ligandability. Interestingly vast majority does keep into consideration kinetics nor thermodynamics devising predictors. This article categorized under: Data Science > Artificial Intelligence/Machine Learning Structure Mechanism Computational Biochemistry Biophysics Chemoinformatics

Language: Английский

---

AF2BIND: Predicting ligand-binding sites using the pair representation of AlphaFold2

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Oct. 18, 2023

Abstract Predicting ligand-binding sites, particularly in the absence of previously resolved homologous structures, presents a significant challenge structural biology. Here, we leverage internal pairwise representation AlphaFold2 (AF2) to train model, AF2BIND, accurately predict small-molecule-binding residues given only target protein. AF2BIND uses 20 “bait” amino acids optimally extract binding signal small-molecule ligand. We find that AF2 pair outperforms other neural-network representations for binding-site prediction. Moreover, unique combinations bait are correlated with chemical properties

Language: Английский

---

A Point Cloud Graph Neural Network for Protein–Ligand Binding Site Prediction

International Journal of Molecular Sciences, Journal Year: 2024, Volume and Issue: 25(17), P. 9280 - 9280

Published: Aug. 27, 2024

Predicting protein-ligand binding sites is an integral part of structural biology and drug design. A comprehensive understanding these essential for advancing innovation, elucidating mechanisms biological function, exploring the nature disease. However, accurately identifying remains a challenging task. To address this, we propose PGpocket, geometric deep learning-based framework to improve site prediction. Initially, protein surface converted into point cloud, then chemical properties each are calculated. Subsequently, cloud graph constructed based on inter-point distances, neural network (GNN) applied extract analyze information predict potential sites. PGpocket trained scPDB dataset, its performance verified two independent test sets, Coach420 HOLO4K. The results show that achieves 58% success rate dataset 56% HOLO4K dataset. These surpass competing algorithms, demonstrating PGpocket's advancement practicality

Language: Английский

---

Comparative evaluation of methods for the prediction of protein–ligand binding sites

Journal of Cheminformatics, Journal Year: 2024, Volume and Issue: 16(1)

Published: Nov. 11, 2024

The accurate identification of protein-ligand binding sites is critical importance in understanding and modulating protein function. Accordingly, ligand site prediction has remained a research focus for over three decades with 50 methods developed change paradigm from geometry-based to machine learning. In this work, we collate 13 predictors, spanning 30 years, focusing on the latest learning-based such as VN-EGNN, IF-SitePred, GrASP, PUResNet, DeepPocket compare them established P2Rank, PRANK fpocket earlier like PocketFinder, Ligsite Surfnet. We benchmark against human subset our new curated reference dataset, LIGYSIS. LIGYSIS comprehensive complex dataset comprising 30,000 proteins bound ligands which aggregates biologically relevant unique interfaces across biological units multiple structures same protein. an improvement testing datasets sc-PDB, PDBbind, MOAD, COACH420 HOLO4K either include 1:1 complexes or consider asymmetric units. Re-scoring predictions by display highest recall (60%) whilst IF-SitePred presents lowest (39%). demonstrate detrimental effect that redundant performance well beneficial impact stronger pocket scoring schemes, improvements up 14% (IF-SitePred) 30% precision (Surfnet). Finally, propose top-N+2 universal metric urge authors share not only source code their methods, but also benchmark.Scientific contributionsThis study conducts largest date, comparing original 15 variants using 10 informative metrics. introduced, highlights demonstrates significant through schemes. proposed prediction, recommendation open-source sharing both benchmarks.

Language: Английский

---

Improving Identification of Drug-Target Binding Sites Based on Structures of Targets Using Residual Graph Transformer Network

Biomolecules, Journal Year: 2025, Volume and Issue: 15(2), P. 221 - 221

Published: Feb. 3, 2025

Improving identification of drug-target binding sites can significantly aid in drug screening and design, thereby accelerating the development process. However, due to challenges such as insufficient fusion multimodal information from targets imbalanced datasets, enhancing performance prediction models remains exceptionally difficult. Leveraging structures targets, we proposed a novel deep learning framework, RGTsite, which employed Residual Graph Transformer Network improve sites. First, residual 1D convolutional neural network (1D-CNN) pre-trained model ProtT5 were extract local global sequence features target, respectively. These then combined with physicochemical properties amino acid residues serve vertex graph. Next, edge incorporated, graph transformer (GTN) was applied more comprehensive features. Finally, fully connected used classify whether site. Experimental results showed that RGTsite outperformed existing state-of-the-art methods key evaluation metrics, F1-score (F1) Matthews Correlation Coefficient (MCC), across multiple benchmark datasets. Additionally, conducted interpretability analysis for through real-world cases, confirmed effectively identify practical applications.

Language: Английский

---

Recent advances in AI-driven protein-ligand interaction predictions

Current Opinion in Structural Biology, Journal Year: 2025, Volume and Issue: 92, P. 103020 - 103020

Published: Feb. 24, 2025

Language: Английский

---

Skittles: GNN‐Assisted Pseudo‐Ligands Generation and Its Application for Binding Sites Classification and Affinity Prediction

Proteins Structure Function and Bioinformatics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 28, 2025

ABSTRACT Nowadays, multiple solutions are known for identifying ligand–protein binding sites. Another important task is labeling each point of a site with the appropriate atom type, process as pseudo‐ligand generation. The number generation limited, and, to our knowledge, influence machine learning techniques has not been studied previously. Here, we describe Skittles, new graph neural network‐assisted approach, and compare it force‐field‐based methods. We also demonstrate application Skittles‐based data solving several problems in structural biology, including classification affinity prediction.

Language: Английский

---

bindNode24: Competitive binding residue prediction with 60% smaller model

Computational and Structural Biotechnology Journal, Journal Year: 2025, Volume and Issue: 27, P. 1060 - 1066

Published: Jan. 1, 2025

Many proteins function through ligand binding. Yet, reliable experimental binding data remains limited. Recent advances predict residues from sequences using protein Language Model embeddings. The AlphaFold Protein Structure Database, which has 3D structure predictions AlphaFold2, opens the way for graph neural networks that residues. Here, we introduce bindNode24, a new method Graph Neural Networks to whether residue binds any of three classes: small molecules, metal ions, and nucleic macromolecules. Compared state-of-the-art, this approach reduces number free parameters by almost 60 % at similar performance. Our findings also suggest secondary tertiary features AlphaFold2 are easy integrate into prediction tasks previously solely relied on

Language: Английский

---