Explainable Graphical Model-Based Transcriptomic Clock from Single Cell Gene Expression Data

Abstract

Predicting the biological age of an organism is a critical task, and multiple accurate models currently exist. However, to extend both lifespan and more critically, healthspan, it is essential to identify points of intervention and understand how to address them. Effective treatments, whether through medication or other methods, require the ability to diagnose the underlying causes of reduced lifespan, like the up- or down-regulation of certain genes. In this study, we evaluate the performance of a modern structure-learning algorithm in constructing a graphical model-based clock for predicting and analyzing biological age. Using a single-cell gene expression atlas, we develop a probabilistic graphical model capable not only of predicting biological age but also of identifying potential gene-level interventions. We compare the performance of various Bayesian and non-Bayesian algorithms for age prediction, assessing them based on their predictive accuracy and their ability to elucidate complex biological processes associated with aging in Caenorhabditis elegans.