Modeling transcriptomic age using knowledge-primed artificial neural networks

This paper introduces a novel age clock – a machine learning model predicting age from biological data – that combines high predictive accuracy with interpretability. Unlike previous "black box" models, this artificial neural network incorporates prior knowledge of biological pathways into its architecture. Using gene expression data from skin tissue, the model accurately predicts age while simultaneously revealing the aging states of the pathways involved. The model's utility is demonstrated by its ability to recapitulate known aging gene associations and decipher the effects of accelerated aging conditions (like Hutchinson-Gilford progeria syndrome) and longevity interventions (like caloric restriction).

Nicholas Holzscheck, Cassandra Falckenhayn, Jörn Söhle, Boris Kristof, Ralf Siegner, André Werner, Janka Schössow, Clemens Jürgens, Henry Völzke, Horst Wenck, Marc Winnefeld, Elke Grönniger, Lars Kaderali