L3-PPI: A Model-Agnostic Approach for Protein-Protein Interaction Prediction
Researchers have introduced L3-PPI, a novel model-agnostic method designed to enhance protein-protein interaction (PPI) prediction. While current learning-based predictors focus heavily on generating powerful protein representations, they often neglect specialized classification heads, relying instead on generic aggregation methods like concatenation or dot products that lack biological insight. Addressing this gap, the study leverages the biological 'L3 rule,' which posits that multiple length-3 paths between protein pairs indicate a higher likelihood of interaction. The proposed L3-PPI method employs an L3-path-regularized graph prompt learning technique to generate prompt graphs with virtual L3 paths based on protein representations. This approach reformulates the classification of protein embedding pairs into a graph-level classification task. As a lightweight, plug-and-play module, L3-PPI seamlessly integrates with existing PPI predictors, injecting interaction priors of complementarity to boost performance. Empirical evidence from popular PPI datasets supports the L3 rule, and extensive experiments demonstrate that L3-PPI achieves superior performance enhancements compared to advanced competitors, offering a biologically informed solution to improve accuracy in cellular function and disease mechanism studies.
Wire timeline
L3-PPI: A Model-Agnostic Approach for Protein-Protein Interaction Prediction
Researchers have introduced L3-PPI, a novel model-agnostic method designed to enhance protein-protein interaction (PPI) prediction. While current learning-based predictors focus heavily on generating powerful protein representations, they often neglect specialized classification heads, relying instead on generic aggregation methods like concatenation or dot products that lack biological insight. Addressing this gap, the study leverages the biological 'L3 rule,' which posits that multiple length-3 paths between protein pairs indicate a higher likelihood of interaction. The proposed L3-PPI method employs an L3-path-regularized graph prompt learning technique to generate prompt graphs with virtual L3 paths based on protein representations. This approach reformulates the classification of protein embedding pairs into a graph-level classification task. As a lightweight, plug-and-play module, L3-PPI seamlessly integrates with existing PPI predictors, injecting interaction priors of complementarity to boost performance. Empirical evidence from popular PPI datasets supports the L3 rule, and extensive experiments demonstrate that L3-PPI achieves superior performance enhancements compared to advanced competitors, offering a biologically informed solution to improve accuracy in cellular function and disease mechanism studies.
cs.AI updates on arXiv.org