Multi-Granularity Cross-Attention Mixture-of-Experts for Protein–Ligand Dissociation Kinetics Prediction

MOTIVATION: Drug-target affinity prediction is pivotal in drug discovery. However, most existing computational methods focus primarily on binding affinity and do not explicitly model dissociation kinetics (Koff). Accurately predicting Koff therefore remains both important and challenging for kinetics-guided drug screening and residence-time optimization. RESULTS: To address this challenge, we propose MGCA, a multi-granularity cross-attention framework for Koff prediction. MGCA represents proteins using multi-depth ESM2 features and ligands using multi-radius Morgan fingerprints. It integrates these representations through gated expert fusion, low-rank bidirectional cross-attention, and protein-anchored residual fusion. Across the Zhao dissociation-kinetics dataset and the KinetX benchmark, MGCA achieved the lowest mean squared error (MSE) in all warm-start and protein cold-start evaluations. Notably, MGCA reduced protein cold-start MSE by 42.5% and 12.1% relative to the recent kinetics-specific BiCoA-Net model on Zhao and KinetX, respectively. Under matched hardware, MGCA also trained substantially faster than BiCoA-Net and MoE-Kinetic. Ablation and case-study analyses further showed that MGCA effectively integrates protein, ligand, and interaction information, supporting cross-target generalization, within-target compound ranking, and feature-level interpretation. Together, these results establish MGCA as an accurate, efficient, and interpretable framework for protein-ligand dissociation kinetics prediction. AVAILABILITY AND IMPLEMENTATION: The source code and datasets are available at https://github.com/CSUBioGroup/MGCA-koff and https://doi.org/10.5281/zenodo.22932383.

Authors

Institutions

Publication Details

Journal
Bioinformatics
Published
2026-11-18
DOI
https://doi.org/10.1093/bioinformatics/btag843
Primary Topic
Computational Drug Discovery Methods
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multi-Granularity Cross-Attention Mixture-of-Experts for Protein–Ligand Dissociation Kinetics Prediction

Gang Luo, Zhangli Lu, Min Li, Long Wen et al.
4 links
Bioinformatics
Computational Drug Discovery Methods
article

Multi-Granularity Cross-Attention Mixture-of-Experts for Protein–Ligand Dissociation Kinetics Prediction

Gang Luo, Zhangli Lu, Min Li, Long Wen, Ziying Cao
article en

Abstract

MOTIVATION: Drug-target affinity prediction is pivotal in drug discovery. However, most existing computational methods focus primarily on binding affinity and do not explicitly model dissociation kinetics (Koff). Accurately predicting Koff therefore remains both important and challenging for kinetics-guided drug screening and residence-time optimization. RESULTS: To address this challenge, we propose MGCA, a multi-granularity cross-attention framework for Koff prediction. MGCA represents proteins using multi-depth ESM2 features and ligands using multi-radius Morgan fingerprints. It integrates these representations through gated expert fusion, low-rank bidirectional cross-attention, and protein-anchored residual fusion. Across the Zhao dissociation-kinetics dataset and the KinetX benchmark, MGCA achieved the lowest mean squared error (MSE) in all warm-start and protein cold-start evaluations. Notably, MGCA reduced protein cold-start MSE by 42.5% and 12.1% relative to the recent kinetics-specific BiCoA-Net model on Zhao and KinetX, respectively. Under matched hardware, MGCA also trained substantially faster than BiCoA-Net and MoE-Kinetic. Ablation and case-study analyses further showed that MGCA effectively integrates protein, ligand, and interaction information, supporting cross-target generalization, within-target compound ranking, and feature-level interpretation. Together, these results establish MGCA as an accurate, efficient, and interpretable framework for protein-ligand dissociation kinetics prediction. AVAILABILITY AND IMPLEMENTATION: The source code and datasets are available at https://github.com/CSUBioGroup/MGCA-koff and https://doi.org/10.5281/zenodo.22932383.

Bioinformatics
Central South University (CN)
Computational Drug Discovery Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon