Subnetwork Mining Based on Structure Entropy for Analyzing Depression Disease

The exponential growth in protein-protein interaction network provides lots of opportunities and challenges to identify un-annotated essential proteins and reveal functional modules associated with depression. Mining associated subnetworks is an effective method to explore depression from a system-level, since it can reduce the size of network and make network analysis feasible and more focused. In this paper, we propose a subnetwork mining method to extract disease associated subnetworks by minimizing structure entropy that highlights the influence of key proteins and functional modules on the disease in question. Based on protein-protein interaction network and known key proteins, our approach initiates by constructing an initial subnetwork induced by a set of known key proteins. Subsequently, we partition this subnetwork into distinct modules using a structure entropy minimization algorithm. Each of these modules within the subnetwork can be considered a potential protein complex, allowing us to uncover their functional implications. Finally, we selectively extract a set of proteins by minimizing structure entropy, thereby ensuring the acquisition of relatively complete modules that better represent the functional aspects of the system. To reduce computational cost, we explore the relationship between the increase of structure entropy and the variance of the edge number sequence, which comprises the counts of edges connecting newly added nodes to existing modules. As a result, we transform the task from minimizing structure entropy to the more tractable problem of maximizing the variance of the edge number sequence. This approach optimizes our computational efficiency while retaining the effectiveness of our method. We evaluate the approach on tasks related to depression. Based on the extracted subnetwork, our model exhibited strong performance in predicting essential proteins across 9 centrality measurements. We selected 23 newly discovered proteins with degree centrality values exceeding 0.2 for enrichment analysis, the enriched pathways primarily pertained to stress-activated MAPK cascade, positive regulation of DNA-binding transcription factor activity, and regulation of inflammatory response etc., which highly related to depression. We also evaluate biological significance of functional modules on obtained subnetwork. The KEGG pathway enrichment analysis reveal that the six modules with the highest number of associated pathways, which relating to signaling pathways, neurodegenerative diseases and nervous system, among others. Furthermore, KEGG pathway enrichment analysis conducted on the predicted proteins and functional modules underscored the rationality and validity of our algorithm. This approach provides a fresh perspective for exploring the pathological mechanisms underlying depression.