Gene Interactions in Survival Data Analysis: A Data-Driven Approach Using Restricted Mean Survival Time and Literature Mining.

Unveiling gene interactions is crucial for comprehending biological processes, particularly their combined impact on phenotypes. Computational methodologies for gene interaction discovery have been extensively studied, but their application to censored data has yet to be thoroughly explored. Our work introduces a data-driven approach to identifying gene interactions that profoundly influence survival rates through the use of survival analysis. Our approach calculates the restricted mean survival time (RMST) for gene pairs and compares it against their individual expressions. If the interaction’s RMST exceeds that of the individual gene expressions, it suggests a potential functional association. We focused on L1000 landmark genes using TCGA na METABRIC data sets. Our findings demonstrate numerous additive and competing interactions and a scarcity of XOR-type interactions. We substantiated our results by cross-referencing with existing interactions in STRING and BioGRID databases and using large language models to summarize complex biological data. Although many potential gene interactions were hypothesized, only a fraction have been experimentally explored. This novel approach enables biologists to initiate a further investigation based on our ranked gene pairs and the generated literature summaries, thus offering a comprehensive, data-driven approach to understanding gene interactions affecting survival rates.