The development of ‘age clocks’, machine learning models predicting aging state from biological data, has been a major innovation in aging research in recent years. Despite their popularity however, most of these models perform essentially as ‘black boxes’, offering little insight into the underlying processes. We propose a new type of clock, one that couples predictivity with interpretability of the underlying biology, achieved through the incorporation of prior knowledge into the model design. The clock, an artificial neural network constructed according to well-described biological pathways, allows the prediction of biological age from gene expression data of epidermal skin tissue with high accuracy, while at the same time capturing and revealing aging states of the pathways driving the predictions. Simulation experiments using the trained model both recapitulate known associations of age related genes such as SIRT1 and TXNIP, as well identify new target genes such as HK2 and KLF4, showcasing the value of interpretable machine learning models for skin aging research. The model further demonstrates its utility in deciphering the main pathways by which accelerated aging conditions such as Hutchinson-Gilford progeria syndrome and photoaging exert their effects.

ESDRLB006 - An interpretable neural age clock for skin tissue

50th Annual Meeting of the European Society for Dermatological Research. VIRTUAL MEETING