Deep Learning of Cellular Metabolic Flux Distributions Predicts Lifespan

Area of Science:

• * Systems biology and computational biology
• * Cellular aging and longevity research
• * Metabolic network analysis

Background:

• * Aging rates vary within species, even among genetically identical individuals in uniform environments.
• * This suggests intrinsic biological mechanisms beyond genetics and environment drive aging.
• * Understanding lifespan variance is crucial for aging research.

Purpose of the Study:

• * To investigate the fundamental causes of lifespan variance in *Saccharomyces cerevisiae*.
• * To determine if metabolic network flux distributions can predict replicative lifespan.
• * To identify key metabolic pathways influencing aging rate.

Main Methods:

• * Utilized a comprehensive metabolic model (yeast GEM_v8.5.0) of *S. cerevisiae*.
• * Simulated gene knockouts to analyze reaction flux space and generated 406,500 flux distributions.
• * Collected replicative lifespan (RLS) data from 66,400 individual cells across 812 viable mutants.
• * Employed Principal Component Analysis (PCA) and deep learning (RNN, CfNN, CNN) to correlate flux configurations with RLS.

Main Results:

• * Identified a core network of highly correlated metabolic reactions that control aging rate.
• * Demonstrated that metabolic flux configurations sufficiently explain all observed lifespan variance.
• * Highlighted specific biosynthetic pathways (ceramides, glycerolipids, etc.) as critical for lifespan regulation.

Conclusions:

• * Replicative lifespan variance in *S. cerevisiae* is an emergent property of its metabolic network.
• * Metabolic flux configurations converge towards three stable states: extended, shortened, and wild-type lifespan.
• * Metabolic network dynamics provide a fundamental explanation for aging rate variation.