Large Language Models as Tools for Molecular Toxicity Prediction: AI Insights into Cardiotoxicity

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 21, 2025

The importance of drug toxicity assessment lies in ensuring the safety and efficacy pharmaceutical compounds. Predicting is crucial development risk assessment. This study compares performance GPT-4 GPT-4o with traditional deep-learning machine-learning models, WeaveGNN, MorganFP-MLP, SVC, KNN, predicting molecular toxicity, focusing on bone, neuro, reproductive toxicity. results indicate that comparable to models certain areas. We utilized combined docking techniques cardiotoxicity three specific targets, examining Chinese medicinal materials listed as both food medicine. approach aimed explore potential mechanisms action. found components Black Sesame, Ginger, Perilla, Sichuan Pagoda Tree Fruit, Galangal, Turmeric, Licorice, Yam, Amla, Nutmeg exhibit toxic effects cardiac target Cav1.2. indicated significant binding affinities, supporting hypothesis cardiotoxic effects.This research highlights ChatGPT properties its significance chemistry, demonstrating facilitation a new paradigm: data set, high-accuracy learning can be generated without requiring computational knowledge or coding skills, making it accessible easy use.

Language: Английский

Semisupervised Learning to Boost hERG, Nav1.5, and Cav1.2 Cardiac Ion Channel Toxicity Prediction by Mining a Large Unlabeled Small Molecule Data Set

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(16), P. 6410 - 6420

Published: Aug. 7, 2024

Predicting drug toxicity is a critical aspect of ensuring patient safety during the design process. Although conventional machine learning techniques have shown some success in this field, scarcity annotated data poses significant challenge enhancing models' performance. In study, we explore potential leveraging large unlabeled small molecule sets using semisupervised to improve cardiotoxicity predictive performance across three cardiac ion channel targets: voltage-gated potassium (hERG), sodium (Nav1.5), and calcium (Cav1.2). We extensively mined ChEMBL database, comprising approximately 2 million molecules, then employed construct robust classification models for purpose. achieved boost on highly diverse (i.e., structurally dissimilar) test all targets. Using our built models, screened whole database set FDA-approved drugs, identifying several compounds with activity. To ensure broad accessibility usability both technical nontechnical users, developed cross-platform graphical user interface that allows users make predictions gain insights into drugs other molecules. The software made available as open source under permissive MIT license at https://github.com/issararab/CToxPred2.

Language: Английский

T-ALPHA: A Hierarchical Transformer-Based Deep Neural Network for Protein-Ligand Binding Affinity Prediction With Uncertainty-Aware Self-Learning for Protein-Specific Alignment

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 20, 2024

Abstract There is significant interest in targeting disease-causing proteins with small molecule inhibitors to restore healthy cellular states. The ability accurately predict the binding affinity of molecules a protein target silico enables rapid identification candidate and facilitates optimization on-target potency. In this work, we present T-ALPHA, novel deep learning model that enhances protein-ligand prediction by integrating multimodal feature representations within hierarchical transformer framework capture information critical predicting affinity. T-ALPHA outperforms all existing models reported literature on multiple benchmarks designed evaluate scoring functions. Remarkably, maintains state-of-the-art performance when utilizing predicted structures rather than crystal structures, powerful capability real-world drug discovery applications where experimentally determined are often unavailable or incomplete. Additionally, an uncertainty-aware self-learning method for protein-specific alignment does not require additional experimental data, demonstrate it improves T-ALPHA’s rank compounds biologically targets such as SARS-CoV-2 main protease epidermal growth factor receptor. To facilitate implementation reproducibility results presented paper, have made our software available at https://github.com/gregory-kyro/T-ALPHA .

Language: Английский