AI Insight
This study introduces EquiPocket, a geometric graph neural network designed to predict protein binding sites for drug discovery applications. The method addresses limitations of existing CNN-based approaches by using an E(3)-equivariant architecture that better represents irregular protein structures, maintains rotation invariance, and effectively characterizes protein surfaces through three specialized modules that extract local geometry, model chemical and spatial structure, and capture surface geometry via message passing. Experimental results on multiple benchmarks demonstrate superior performance compared to current state-of-the-art methods.
Why it matters
Accurate prediction of protein binding sites is critical for accelerating drug discovery and development. This improved computational approach could reduce the time and cost associated with identifying potential drug targets and designing therapeutic compounds by providing more reliable predictions that account for protein structural properties.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Predicting the binding sites of target proteins plays a fundamental role in drug discovery. Most existing deep-learning methods consider a protein as a 3D image by spatially clustering its atoms into voxels and then feed the voxelized protein into a 3D CNN for prediction. However, the CNN-based methods encounter several critical issues: 1) defective in representing irregular protein structures; 2) sensitive to rotations; 3) insufficient to characterize the protein surface; 4) unaware of protein size shift. To address the above issues, this work proposes EquiPocket, an E(3)-equivariant Graph Neural Network (GNN) for binding site prediction, which comprises three modules: the first one to extract local geometric information for each surface atom, the second one to model both the chemical and spatial structure of protein and the last one to capture the geometry of the surface via equivariant message passing over the surface atoms. We further propose a dense attention output layer to alleviate the effect incurred by variable protein size. Extensive experiments on several representative benchmarks demonstrate the superiority of our framework to the state-of-the-art methods.
Source: EquiPocket: an E(3)-Equivariant Geometric Graph Neural Network for Ligand Binding Site Prediction