Toward Metagenome-Scale Metabolic Flux and Free Energy Analysis via Deep Learning
Junyoung O. Park, University of California, Los Angeles (Principal Investigator)
Pin-Kuang Lai, Stevens Institute of Technology (Principal Investigator)

Metabolism is a dynamic network of biochemical reactions that supply cellular building blocks and energy. Precisely controlling metabolic pathways to redirect biochemical and energetic resources would enable efficient and sustainable production of advanced biofuels and bioproducts. However, challenges arise from the lack of the ability to quantify the rates at which metabolic pathways are utilized. In this project, the investigators develop a computational toolset for genome-scale and metagenome-scale quantification of metabolic pathway utilization and the thermodynamic driving force behind it. The investigators combine deep learning with stable isotope tracing and simulation techniques. Stable isotope tracers imprint characteristic labeling patterns corresponding to metabolic activity on cellular molecules upon entering the cell. Using multilayer neural networks, deep learning models predict metabolic rates and thermodynamic energies from the isotope labeling patterns of metabolites in large complex biological systems such as microbial communities. With the augmented quantitative capability, the new software will i) interpret metabolomic and other omic data coherently, ii) impart quantitative systems-level knowledge of metabolism in individual and across multiple species, and iii) reveal precise metabolic control strategies. The resulting tool and knowledge offer dual benefits of laying a solid foundation for metabolic engineering and integrating large-scale biochemical datasets.