(368bt) Systems Biology Approaches for Engineering Metabolism Using Isotope Tracing and Machine Learning

Metabolic fluxes are a fundamental descriptor of cellular state, representing the rates at which organisms operate metabolic pathways. Mass spectrometry and isotope tracing have been instrumental in quantifying fluxes, as metabolic pathways imprint unique isotope labeling patterns on metabolites corresponding to their fluxes. Metabolic flux analysis (MFA) is a commonly used computational framework that identifies the set of fluxes that best simulate observed isotope labeling patterns. However, quantitative flux analysis remains an expert method, and the relationships between isotopic labeling patterns and fluxes remain elusive in complex metabolic environments. Here, we aimed to make flux quantitation tools more scalable and accessible by innovating a two-stage machine learning (ML) framework termed ML-Flux. ML-Flux is trained using data from five universal models of central carbon metabolism and 26 different ¹³C and ²H glucose and glutamine tracers to convert isotope labeling patterns into metabolic fluxes. Using ML-Flux with multi-isotope tracing, we determined metabolic fluxes through central carbon metabolism at orders-of-magnitude faster speeds than traditional MFA. Additionally, ML-Flux computes reaction Gibbs free energies, which informs on the efficiency of energy or enzyme usage in reaction steps. The ML-Flux analysis frameworks is computationally light and deployed as a webtool. Thus, ML-assisted isotope tracing is a promising step towards democratizing flux quantitation in increasingly complex biological systems (i.e. genome-scale models and cellular communities), thereby expanding our understanding and control of metabolism.