Core F EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR AGING

CORE F: EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR AGING – PROJECT SUMMARY This Core aims to provide the biology of aging research community with access to cutting-edge artificial intelligence (AI) technologies through the University of Washington Nathan Shock Center (NSC). The Explainable AI Core focuses on developing AI methods for analyzing high-dimensional omics data to understand complex phenotypes. Its primary objective is to extract biologically meaningful insights from multi- omic datasets to assist geroscience researchers in interpreting these datasets effectively and to contribute to the development of new AI techniques tailored for aging research. The Core emphasizes Explainable AI (XAI) to improve the interpretability of complex machine learning models. With the increase in multi-omic datasets from tissues of varying ages, applying machine learning models to learn low-dimensional embeddings is crucial. Deep neural networks and other complex models are essential for AI applications but often lack transparency. Therefore, Core F aims to enhance the use of robust XAI methods in aging biology research and develop novel XAI approaches to actively translate model explanations into actionable biological insights and mechanistic hypotheses. It will collaborate with other Cores and the wider aging research community to explore the molecular basis of aging and support the adoption of novel AI tools. In Aim 1 we will develop and apply XAI methods for biologically interpretable embeddings. These will provide state-of-the-art XAI techniques for unsupervised embedding models to facilitate data-driven discoveries in aging biology from unimodal or multimodal high-dimensional datasets. In Aim 2 we focus on designing XAI techniques to identify aging biomarkers from NSC user generated or publicly available databases. These efforts will ensure accurate computation of feature attributions using mathematically grounded approaches like Shapley values to identify biologically accurate biomarkers that are connected to mechanistic information on the drivers of aging. In Aim 3 we apply XAI approaches to construct next-generation biological aging clocks. This work will build on the success of the Explainable Biological Age (ENABL Age) model to create accurate and interpretable models that incorporate diverse features, outcomes, and advanced XAI techniques. The XAI methods of Core F provide a point of integration for the UW NSC Resource Cores whereby proteomic and metabolomic data are used to generate specific mechanistic hypothesis that can be tested using gene editing and phenotyping approaches using invertebrate models in Core E. These XAI approaches will facilitate robust model explanations and accelerate novel biological discoveries in aging biology, ultimately contributing to the development of personalized interventions for healthy aging. Additionally, these XAI approaches are generalizable across various geroscience research contexts and biological research fields.