GenShin:geometry-enhanced structural graph embodies binding pose can better predicting compound-protein interaction affinity

TL;DR

This work introduces GenShin, a geometry-enhanced structural-graph model for predicting compound–protein affinity without requiring precise docking poses. By separately encoding protein and compound geometry with dual encoders and coupling them through an interaction module and a dual-loss predictor, GenShin learns from a ground-truth distance map to implicitly encode docking conformation. It achieves state-of-the-art results among pose-free methods on the PDBbind-v2020 benchmark and remains competitive with pose-dependent models on PDBbind-v2016 and CASF-2016, while exhibiting robustness to inadequate docking poses. The approach also enables reconstruction of docking coordinates from predicted distance maps, suggesting practical utility for high-throughput virtual screening in real-world drug discovery. Overall, GenShin offers a scalable, pose-free pathway to accurate affinity prediction with tangible implications for accelerating structure-based drug design.

Abstract

AI-powered drug discovery typically relies on the successful prediction of compound-protein interactions, which are pivotal for the evaluation of designed compound molecules in structure-based drug design and represent a core challenge in the field. However, accurately predicting compound-protein affinity via regression models usually requires adequate-binding pose, which are derived from costly and complex experimental methods or time-consuming simulations with docking software. In response, we have introduced the GenShin model, which constructs a geometry-enhanced structural graph module that separately extracts additional features from proteins and compounds. Consequently, it attains an accuracy on par with mainstream models in predicting compound-protein affinities, while eliminating the need for adequate-binding pose as input. Our experimental findings demonstrate that the GenShin model vastly outperforms other models that rely on non-input docking conformations, achieving, or in some cases even exceeding, the performance of those requiring adequate-binding pose. Further experiments indicate that our GenShin model is more robust to inadequate-binding pose, affirming its higher suitability for real-world drug discovery scenarios. We hope our work will inspire more endeavors to bridge the gap between AI models and practical drug discovery challenges.

Figures (3)

• Figure 1: Overview of GenShin Model.
• Figure 2: Visual comparisons of docking conformation after rotations.
• Figure 3: Visual comparisons of docking conformation reconstructed from dismap GenShin predicts