AI-Generated Hallmarks of Aging and Cancer: A Computational Approach Using Causal Emergence and Dependency Networks

Abstract

This study introduces "hallmarks engineering," a computational approach to generate quantifiable hallmarks of aging and cancer. We evaluated these hallmarks using genome-wide DNA methylation data from ten age-related diseases. Causal emergence analysis revealed that hallmark-level features show stronger disease associations than individual genes, with improvements up to 9.7 orders of magnitude. Hallmark-based models achieved comparable predictive performance with fewer predictors compared to regular pathway-based models. Dependency network analysis uncovered regulatory networks with power-law distributions and identified top-level "super-regulators" such as genomic stability. Notably, the inclusion of neurodegenerative and cancer hallmarks enhanced representation for their respective disease categories. Our findings suggest that top-down modeling using computationally generated hallmarks may reveal common mechanisms across multiple diseases, offering a promising approach for modeling multimorbidity.