A physics-informed graph neural network to approximate docking-based binding affinity for DYRK2 in Alzheimer’s drug repurposing

2026-02-11 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Mathematics & Computational Sciences, Life Sciences & Biology, Health & Medical Research · Depth: Expert, long

Summary

This study introduces PhysDual-GCN, a physics-informed graph neural network (GNN) designed to rapidly approximate docking-derived binding affinity scores for DYRK2, a target relevant to Alzheimer's disease (AD). Traditional molecular docking methods are computationally expensive, prompting the development of this GNN. PhysDual-GCN processes ligand molecular graphs and a sequence-based graph representation of DYRK2, explicitly integrating Coulomb and Lennard–Jones interaction terms. The model was trained and evaluated using reference labels exclusively from classical docking tools (AutoDock Vina, Smina, QVina, CB-DOCK) due to a lack of experimental data for DYRK2-drug pairs. Despite being trained on a limited dataset of four FDA-approved AD drugs, the model achieved low absolute errors (MAE = 0.31 kcal/mol; RMSE = 0.44 kcal/mol) and correctly identified stronger binders like donepezil (-10.8 kcal/mol) and brexpiprazole (-10.0 kcal/mol). This approach demonstrates enhanced interpretability and computational efficiency compared to classical docking workflows.

Key takeaway

For AI Researchers developing drug discovery tools, PhysDual-GCN offers a blueprint for creating computationally efficient and interpretable models. You should consider integrating explicit physical interaction terms into your GNN architectures to improve transparency and reduce reliance on purely black-box approaches, especially when experimental data is scarce. This method provides a foundation for accelerating early-stage drug repurposing efforts for diseases like Alzheimer's.

Key insights

Integrating physical interaction terms into GNNs can create efficient, interpretable surrogates for computationally intensive molecular docking.

Principles

• Physics-informed GNNs enhance interpretability.
• Computational references can train predictive models.
• Ligand-level separation prevents circularity in model evaluation.

Method

PhysDual-GCN jointly processes ligand molecular graphs and protein sequence-based graphs, incorporating Coulomb and Lennard–Jones terms to approximate docking-derived binding affinities for specific protein targets.

In practice

• Use GNNs for faster binding affinity prediction.
• Incorporate physical terms for model interpretability.
• Utilize classical docking scores as training labels.

Topics

• Physics-informed GNNs
• Drug-target Binding Affinity
• Alzheimer's Drug Repurposing
• Molecular Docking
• DYRK2

Code references

• StarNNT/PhysDual-GCN

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• Physics-Informed Neural Networks for Chemotherapy Pharmacokinetics: Benchmarking the Clinical Estimator and Exposing Parameter Identifiability — PINNs offer a uniform framework for pharmacokinetics, matching traditional methods on linear problems and exposing identifiability issues in nonlinear models.
• Customized Amazon Nova models improve molecular-property prediction in drug discovery — Fine-tuned LLMs can unify complex molecular property prediction, outperforming or matching specialized GNNs.
• Co-Diffusion: An Affinity-Aware Two-Stage Latent Diffusion Framework for Generalizable Drug-Target Affinity Prediction — Co-Diffusion enhances drug-target affinity prediction in cold-start scenarios using a two-stage, affinity-aware latent diffusion framework.
• Physics constrained graph neural network for real time prediction of intracranial aneurysm hemodynamics — A physics-constrained GNN predicts intracranial aneurysm hemodynamics in real-time, generalizing across patient geometries.
• AI is reengineering drug discovery by speeding up testing and scanning petabytes of data for connections between diseases — AI significantly enhances drug discovery by analyzing vast data to predict targets, efficacy, and disease interrelationships, improving success rates.
• OOD-GraphLLM: Graph Large Language Model for Out-of-Distribution Generalized Drug Synergy Prediction — OOD-GraphLLM uses a GraphLLM to predict drug synergy under out-of-distribution conditions by integrating molecular graph and semantic data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.