PersonaAI: An Interactive Agentic-AI Framework for Autonomous Hypothesis Generation and Validation in Aging

Abstract

Elucidating the mechanisms of aging is impeded by its stochastic, multi-scale nature and cellular heterogeneity, challenges that are compounded by the overwhelming volume of biomedical literature and the complexity of genome-wide datasets. To overcome these barriers, we present PersonaAI, an interactive agentic-AI framework that acts as a digital co-scientist. By integrating literature-based reasoning with autonomous in silico validation, PersonaAI synthesizes over 560,000 aging-related publications via retrieval-augmented generation (RAG) to propose mechanistic hypotheses. These hypotheses are subsequently validated by autonomous agents utilizing single-cell RNA-seq data. Using a temporal cutoff strategy restricted to pre-2020 literature, we demonstrate that PersonaAI can generate hypotheses effectively validated by post-2021 discoveries, proving its capacity for inference beyond simple information retrieval. In application, the system identified senescent Cirbp+ hepatocytes as a liver-intrinsic aging program and uncovered a middle-aged, male-specific decline in adipose stem and progenitor cells, driven by vascular niche deterioration and disrupted VEGF-VEGFR signaling. These results establish PersonaAI as a scalable platform that augments human intuition with autonomous data-driven validation, providing a generalizable platform for accelerating discovery in aging biology.