SpatialPPIv2: Enhancing protein–protein interaction prediction through graph neural networks with protein language models
Wenxing Hu, Masahito Ohue
Abstract
Protein-protein interactions (PPIs) are fundamental to cellular functions, and accurately predicting such interactions is crucial for understanding biological mechanisms and facilitating drug discovery. SpatialPPIv2 is an advanced graph-neural-network-based model that predicts PPIs using large language models to embed sequence features and graph attention networks to capture structural information. By leveraging the comprehensive PINDER dataset, which includes interaction data from the RCSB PDB and AlphaFold databases, SpatialPPIv2 improves the PPI prediction specificity and robustness. Unlike the original SpatialPPI, the updated version no longer depends on protein structure prediction algorithms and can predict protein interactions standalone. SpatialPPIv2 outperforms the state-of-the-art PPI predictors, demonstrating superior accuracy and reliability. Furthermore, the model was robust when using structural prediction methods, including AlphaFold3, AlphaFold2, and ESMFold, indicating its applicability even if experimentally determined structures are unavailable. SpatialPPIv2 offers a promising solution for accurately predicting PPIs, provides insight into protein function, and supports advances in drug discovery and synthetic biology. SpatialPPIv2 is available at https://github.com/ohuelab/SpatialPPIv2.