Aging is driven by interconnected genetic, epigenetic, molecular, and physiological processes spanning from unicellular to organismal levels. The surge in high-throughput data, from clinical and imaging to multi-omics, has outpaced traditional analysis methods; driving the integration of artificial intelligence (AI) into aging research. This comprehensive review examines the application of machine learning, deep learning, and computer vision across four canonical aging models (yeast, Caenorhabditis elegans, Drosophila melanogaster, and mice), highlighting AI's role in lifespan prediction, biomarker and gene discovery, aging-clock construction, and assay automation via automated animal counting and imaging. However, only 3% of the reviewed studies incorporated in vivo biological validation with common issues including small and imbalanced datasets, dataset bias, prediction noise, lack of cross-species analyses, absence of cytotoxicity testing, and overreliance on synthetic data. These drawbacks pose AI as just an aiding tool rather than a standalone solution, and without improvements in these sectors, AI-derived findings should be considered hypothesis generating rather than definitive conclusions. To address these issues, we propose the development of a standardized scoring system, AI Quality Assessment Metric (AI-QAM), for aging research that will evaluate studies on six criteria: (1) dataset size, (2) feature dimensionality, (3) biological validation type, (4) species diversity, (5) model generalizability, and (6) interpretability. Moreover, to mitigate the problem of lacking a unifying of a framework integrating AI approaches with biological mechanisms of aging, we present a conceptual framework, mapping AI applications across biological levels and aging hallmarks. AI will fulfill its potential in aging research only when it is firmly grounded in biological principles, systematically benchmarked, and rigorously validated through experimental studies.
Mahbub et al. (Wed,) studied this question.