TF-MOEA$^{+}$: A Multi-Objective Evolutionary Framework With Self-Adaptive Topological-Functional Mutation for Protein Complex Detection.

Abstract

Identifying protein complexes from protein-protein interaction (PPI) networks is a fundamental task in systems biology, offering important insights into cellular organization and molecular function. However, reliable complex detection remains challenging due to the sparsity and noise of PPI data, as well as the need to balance structural compactness, inter-complex separability, and biological coherence. To address these challenges, this study formulates protein complex detection as a tri-objective optimization problem that jointly integrates topological and biological criteria. We propose TF-MOEA$^{+}$, a multi-objective evolutionary framework built on NSGA-II and equipped with two biologically informed search operators. The first is an objective-guided uniform crossover (OG-UX), which biases recombination toward parents with better multi-objective quality while preserving adjacency-feasible inheritance. The second is a self-adaptive topological-functional synergy mutation (TF-SM), which adjusts mutation behavior according to the degree of structural and functional integration of each protein and reassigns weakly integrated proteins using combined topological and GO-based evidence. Comprehensive experiments on four benchmark yeast PPI networks, Yeast-D1, Yeast-D2, Collins-CYC2008, and Collins-MIPS, show that TF-MOEA$^{+}$ consistently outperforms all baseline methods considered in this study in both detection accuracy and functional relevance. Additional ablation, statistical significance, sensitivity, and runtime analyses further demonstrate the robustness and efficiency of the proposed framework.