Aging is the greatest risk factor for chronic diseases, premature death, morbidity and disability. With significant variability in the trajectory of biological aging (BAg) among individuals of the same chronological age and human tissues, biomarker of BAg may offer better estimate of disease risk and mortality than chronological age. A promising candidate is plasma aging-related proteins (AP), which originate from nearly every organ and cell type. A recent study has demonstrated that a proteomic aging score can be estimated from 204 plasma proteins, which can predict mortality and is closely associated with the risk of 18 major chronic diseases. Another study has shown that BAg of individual’s organ system can be estimated from the corresponding organ system’s phenotypes that are routinely assayed in primary care setting, or readily assayed at minimal cost. These “organ clocks” are associated with the risk of 16 chronic diseases. The inter-dependence of aging across organs is also highlighted by the influence of the biological age (BA) of one organ on that of another organ system. Recognizing the importance of plasma AP in elucidating aging biology and the inter-relationship between the BA of organ systems, we hypothesize that a model that predicts organ-specific BA based on whole-body phenotypes and plasma AP would outperform one that is built on either alone. To this end, we aim to develop a deep learning model to predict the BA of individual organ system using routinely or readily assayed phenotypes, while also accounting for plasma AP. We place a heavy emphasis on the practicality of model deployment. We aim to identify the minimal subsets of plasma AP and phenotypes that (i) are crucial for predicting BA, and (ii) yield robust, clinically actionable prediction accuracies. We will investigate the relationship between predicted BA versus the risk of chronic diseases and mortality, and the trade-offs between model complexity and prediction performance.
