Identifying Protein Complexes in Protein-Protein Interaction Data Using Graph Convolutional Network

Protein complexes are groups of two or more polypeptide chains that bind to form noncovalent networks protein interactions. Over the past decade, researchers have created a number means computing ways in which and their members can be identified through these interaction networks. Although most existing methods identify functional from protein-protein (PPIs) at fairly decent level, applicability advanced graph network has not yet been adequately investigated. This paper proposes various convolutional (GCN) improve detection complexes. We first formulate complex problem as node classification problem. Then, we developed Neural Overlapping Community Detection (NOCD) model cluster nodes (proteins) using affiliation matrix. A representation learning approach, combines multi-class GCN feature extractor (to obtain nodes’ features) mean shift clustering algorithm perform clustering), is also utilized. convert dense-dense matrix operations into dense-sparse sparse-sparse efficiency by reducing space time complexities. The proposed solution significantly improves scalability GCN. Finally, apply aggregation find best grid search then performed on detected obtained via three well-known methods, namely ClusterONE, CMC, PEWCC, with help Meta-Clustering Algorithm (MCLA) Hybrid Bipartite Graph Formulation (HBGF). test GCN-based publicly available datasets they better than previous state-of-the-art methods. code/data for free download https://github.com/Analystharsh/GCN_complex_detection.