In-silico evaluation of interventions for molecular health/rejuvenation, using omics data analyses by bioinformatics, and generative AI

• Applicant: Professor Dr. Georg Fuellen
• Subject Area: Medical Informatics and Medical Bioinformatics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563142248

Molecular omics data, biomedical background knowledge, and bioinformatics tools may now enable predictions of intervention effects in the areas of biomedicine aiming for longevity, or even for rejuvenation. These predictions must be satisfactory in their correctness, robustness and explainability. AI-based approaches may sooner or later contribute substantially. Good predictions of health efficacy and toxicity based on surrogate biomarkers are needed because in the field of human primary prevention and longevity, clinical studies and medical recommendations can rarely be based on long-term follow-up data, which take decades to collect. Aging clocks are popular as surrogates, but suffer from being trained on observational data, while inferences are much more reliable when based on intervention data; we wish to train directly on these. For interventions in healthy people, explainability/interpretability and safety/toxicity are especially important. We thus develop EIRe, the Evaluator of Interventions of Rejuvenation. EIRe will be an applied medical bioinformatics pipeline that takes as input the gene/protein expression effects of an intervention (or a set of interventions). EIRe places the input samples into similarity space together with other known interventions and suggests biomolecular mechanisms explaining the effects of the interventions, based on gene set enrichment analyses. Its further output is an embedding of the underlying gene-set annotations in their own similarity space, highlighting how the intervention-specific (de-)activations of processes and pathways relate to each other. It focusses on calculating interpretable linear combinations of features, e.g. based on principal component analysis or gene-set enrichment. It will then apply the best machine learning models as predictors for the health efficacy and toxicity of the interventions, specifically assessing the value of the interpretable linear combinations of features. In particular, we will test contrastive PCA that uses the control samples to deal with confounders, and employ automated machine learning (AutoML). Specifically, EiRE shall tell us whether rejuvenation interventions can yield the effects promised in recent years, without toxic side effects. We first use benchmarking data from toxicology studies. We then turn to cellular or organismal rejuvenation, e.g. by small compounds or partial/transient reprogramming. We will obtain and exploit suitable benchmark data, and we will explore ways to improve EIRe results by generative AI. We shall pre-register a major part of our analyses. Overall, with our benchmarking data and the development of EIRe, we aim for simple, sound and explainable ways to evaluate biomedical interventions for health efficacy and toxicity.

• DFG Programme: Research Grants