A Degree Centrality-Enhanced Computational Approach for Local Network Alignment Leveraging Knowledge Graph Embeddings
Abstract
Biological network alignment serves various goals, including identifying similar regions between networks of different species, transferring biological knowledge, and predicting protein complexes and functions. This paper presents KOGAL (KnOwledge Graph ALignment), a scalable multiprocessing algorithm for local protein-protein interaction (PPI) network alignment. KOGAL aims to predict conserved protein complexes across species’ PPI networks. It employs two strategies for seed discovery and initial alignment: the first computes the cosine similarity between embedding vectors derived from knowledge graph models like TransE or DistMult to generate an alignment matrix. The second strategy begins the alignment process by calculating the centrality degree of nodes within each network, highlighting the importance of each protein in the network structure. Protein similarities are quantified by combining protein sequence similarities with knowledge graph embeddings, ensuring biologically meaningful structural alignments. KOGAL depicts strong results compared to other state-of-the-art approaches. When evaluated on real PPI networks, KOGAL demonstrates high accuracy across multiple metrics, including coverage, sensitivity, the number of matched reference conserved complexes (Frac), complex-wise sensitivity (Sn), positive predictive value (PPV), geometric accuracy (ACC), and maximum matching ratio (MMR).