Identifying novel caloric restriction mimetics using generative AI

Abstract

In the contemporary era, anti-aging stands as a prominent research focus. Recognizing the impact of diet on aging, investigations into caloric restriction methods have emerged as a promising and effective approach in anti-aging research, aiming to extend longevity. Despite its potential benefits, a significant drawback of caloric restriction is the necessity for strict lifelong adherence. In response to this limitation, researchers have turned their attention to molecules that mimic the effects of caloric restriction, termed caloric restriction mimetics (CRMs). This study delves into the identification of novel Caloric Restriction Mimetics, employing a Generative AI approach that integrates computational drug discovery and an anti-aging probabilistic model. Specifically, a Graph Neural Network approach and anti-aging score optimization was utilized within the framework of Multi-Constraint Molecule Sampling for Molecule Optimization (MIMOSA). Given the scarcity of existing CRMs, this research generated novel CRMs. By prioritizing bioactivity as a key feature, highly promising novel CRMs were identified through screening, visualization and top-k-neighbour techniques from ZINC20. The outcomes of this analysis yield potential caloric restriction mimetic molecules, paving the way for future research and development in the field. Moreover, the study highlights the applicability of generative modeling approaches in cases where the availability of ground truth data for classical machine learning model building is limited. This research contributes to advancing our understanding of anti-aging strategies and underscores the significance of innovative methodologies in drug discovery within the realm of aging-related interventions.