Protein-Protein Interaction Prediction Based on Information Contribution and Complementarity

Protein-protein interactions play a crucial role in various biological processes. However, only a small fraction of these interactions have been identified, making it a significant challenge to predict potential interactions based on known data. The L3 method pioneered the use of complementarity instead of similarity for link prediction in protein interaction networks and achieved notable results. Nevertheless, it relies solely on third-order paths while overlooking other topological information. To address this limitation, this study proposes a novel link prediction based on information contribution and complementarity(ICC), which integrates protein complementarity with information contribution. First, we introduce the concept of the quadrangle coefficient and apply it to enhance complementarity-based algorithms. Building on this, we define both initial and global information contribution based on network topology. By combining these contributions with complementarity, ICC compensates for the missing topological information, achieving more accurate protein-protein interaction predictions. Experimental results on four datasets, including A. thaliana and C. elegans, demonstrate that ICC outperforms existing methods in terms of precision, AUC, and nDCG metrics.