Topology-Driven Negative Sampling Enhances Generalizability in Protein-Protein Interaction Prediction

Abstract

Unraveling the human interactome to uncover disease-specific patterns and discover drug targets hinges on accurate protein-protein interaction (PPI) predictions. However, challenges persist in machine learning (ML) models due to a scarcity of quality hard negative samples, shortcut learning, and limited generalizability to novel proteins. Here, we introduce a novel approach for strategic sampling of protein-protein non-interactions (PPNIs) by leveraging higher-order network characteristics that capture the inherent complementarity-driven mechanisms of PPIs. Next, we introduce UPNA-PPI (Unsupervised Pre-training of Node Attributes tuned for PPI), a high throughput sequence-to-function ML pipeline, integrating unsupervised pretraining in protein representation learning with topological PPNI samples, capable of efficiently screening billions of interactions. UPNA-PPI improves PPI prediction generalizability and interpretability, particularly in identifying potential binding sites locations on amino acid sequences, strengthening the prioritization of screening assays and facilitating the transferability of ML predictions across protein families and homodimers. UPNA-PPI establishes the foundation for a fundamental negative sampling methodology in graph machine learning by integrating insights from network topology.