Improved supervised prediction of aging-related genes via weighted dynamic network analysis

Abstract

Motivation: This study focuses on supervised prediction of aging-related genes from -omics data. Unlike gene expression methods that capture aging-specific information but study genes in isolation, or protein-protein interaction (PPI) network methods that account for PPIs but the PPIs are contextunspecific, we recently integrated the two data types into an aging-specific PPI subnetwork, which yielded more accurate aging-related gene predictions. However, a dynamic aging-specific subnetwork did improve prediction performance compared to a static aging-specific subnetwork, despite the aging process being dynamic. Results: So, here, we propose computational advances towards improving prediction accuracy from a dynamic aging-specific subnetwork. We develop a supervised learning model that when applied to a dynamic subnetwork yields extremely high prediction performance, with F-score of 91.4%, while the best model on any static subnetwork yields F-score of "only" 74.3%. Hence, our predictive model could guide with high confidence the discovery of novel aging-related gene candidates for future wet lab validation.

Graphlets in multilayer networks

“…predict which group a network or an element of a network belongs in based on the graphlet degrees. Even though such methods have been previously applied to single-layer networks [44,45,46,47], there is a clear avenue for further research in the application of machine learning to multilayer graphlets. Presumably, the problem of choosing the correct set of graphlets is less severe in supervised learning in which the method should be able to find the important orbits based on the training data.…”

DynACPD Embedding Algorithm for Prediction Tasks in Dynamic Networks

“…Recently, matrix factorization methods attempt to formulate different proximity matrices based on the n-hop transitional Also node embedding algorithms in attributed networks [BLM19] and heterogeneous networks[HYCM18] have been explored widely. Despite their success, many real-world scenarios are essentially dynamic, for example, relationships in a social network [SM06], spatio-temporal traffic prediction [ZFWQ20] and progressions of aging related genes [LM20]. Algorithms for static networks generally fail to consider the evolution of network structures, and thus lack the ability to capture any time-dependent information which can affect the performance of downstream tasks.…”