Abstract:Metabolic stability of proteins plays a vital role in various dedicated cellular processes. Traditional methods of measuring the metabolic stability are time-consuming and expensive. Therefore, we developed a more efficient computational approach to understand the protein dynamic action mechanisms in biological process networks. In this study, we collected 341 short-lived proteins and 824 non-short-lived proteins from U2OS; 342 short-lived proteins and 821 non-short-lived proteins from HEK293T; 424 short-lived proteins and 1153 non-short-lived proteins from HCT116; and 384 short-lived proteins and 992 non-short-lived proteins from RPE1. The proteins were encoded by GO and KEGG enrichment scores based on the genes and their neighbors in STRING, resulting in 20,681 GO term features and 297 KEGG pathway features. We also incorporated the protein interaction information from STRING into the features and obtained 19,247 node features. Boruta and mRMR methods were used for feature filtering, and IFS method was used to obtain the best feature subsets and create the models with the highest performance. The present study identified 42 features that did not appear in previous studies and classified them into eight groups according to their functional annotation. By reviewing the literature, we found that the following three functional groups were critical in determining the stability of proteins: synaptic transmission, post-translational modifications, and cell fate determination. These findings may serve as a valuable reference for developing drugs that target protein stability.

Analysis and Prediction of Protein Stability Based on Interaction Network, Gene Ontology, and KEGG Pathway Enrichment Scores.