L3-PPI: Biologically Informed Graph Prompt Learning for Protein-Protein Interaction Prediction
A recent study published on arXiv (2605.09964) presents L3-PPI, a versatile approach for predicting protein-protein interactions (PPI) that leverages insights from the 'L3 rule.' This rule suggests that the presence of multiple length-3 paths between two proteins increases the likelihood of their interaction. The researchers offer empirical support for this rule using well-known PPI datasets. L3-PPI employs L3-path-regularized graph prompt learning to create a prompt graph featuring virtual L3 paths derived from protein representations, while regulating the path count. This method transforms the classification of protein embedding pairs into a graph-centric model, addressing the oversight of specialized classification heads in existing learning-based PPI predictors.
Key facts
- arXiv paper 2605.09964 introduces L3-PPI for PPI prediction.
- L3-PPI is motivated by the biological L3 rule.
- The L3 rule states that multiple length-3 paths between proteins indicate interaction likelihood.
- Empirical evidence shows popular PPI datasets support the L3 rule.
- L3-PPI uses L3-path-regularized graph prompt learning.
- It generates a prompt graph with virtual L3 paths.
- The method controls the number of virtual L3 paths.
- L3-PPI reformulates PPI classification into a graph-based framework.
Entities
Institutions
- arXiv