InteractionGraphNet: A Novel and Efficient Deep Graph Representation Learning Framework for Accurate Protein–Ligand Interaction Predictions

TLDR

Accurate quantification of protein–ligand interactions remains a key challenge to structure‑based drug design, and traditional ML methods using handcrafted descriptors or 2D graph representations are limited in learning generalized 3D interactions. We propose InteractionGraphNet, a deep graph representation learning framework that learns protein–ligand interactions directly from 3D complex structures. IGN stacks two independent graph convolution modules to sequentially learn intramolecular and intermolecular interactions, enabling efficient downstream use of the learned intermolecular features. IGN outperforms or matches state‑of‑the‑art ML baselines and docking programs in binding affinity prediction, virtual screening, and pose prediction, demonstrating that it captures key interaction features rather than memorizing biased patterns.

Abstract

Accurate quantification of protein–ligand interactions remains a key challenge to structure-based drug design. However, traditional machine learning (ML)-based methods based on handcrafted descriptors, one-dimensional protein sequences, and/or two-dimensional graph representations limit their capability to learn the generalized molecular interactions in 3D space. Here, we proposed a novel deep graph representation learning framework named InteractionGraphNet (IGN) to learn the protein–ligand interactions from the 3D structures of protein–ligand complexes. In IGN, two independent graph convolution modules were stacked to sequentially learn the intramolecular and intermolecular interactions, and the learned intermolecular interactions can be efficiently used for subsequent tasks. Extensive binding affinity prediction, large-scale structure-based virtual screening, and pose prediction experiments demonstrated that IGN achieved better or competitive performance against other state-of-the-art ML-based baselines and docking programs. More importantly, such state-of-the-art performance was proven from the successful learning of the key features in protein–ligand interactions instead of just memorizing certain biased patterns from data.