[SS23] Unraveling microbial community dynamics through metabolic interactions

[SS23-03] Machine Learning for Efficient and Robust Simulations of Reactive-Transport Systems Integrated with Microbial Metabolic Networks

*Hyun-Seob Song¹, Firnaaz Ahamed¹, Joon-Yong Lee², Christopher Henry³, Janaka Edirisinghe³, William Nelson², Xingyuan Chen², J. David Moulton⁴, Timothy Scheibe² (1. University of Nebraska-Lincoln (United States of America), 2. Pacific Northwest National Laboratory (United States of America), 3. Argonne National Laboratory (United States of America), 4. Los Alamos National Laboratory (United States of America))

Keywords:

Machine learning,Metabolic networks,Reactive-transport models,Flux balance analysis,Surrogate models

Flux balance analysis (FBA) is widely used to analyze genome-scale metabolic networks, providing valuable insights into microbial metabolism and cellular function. FBA estimates metabolic flux distributions by solving a linear programming (LP) problem to optimize a predefined objective, such as biomass production. There is growing interest in integrating metabolic networks with reactive-transport models (RTMs) to better understand the complex interplay between microbial metabolism and the spatiotemporal movement of fluids and solutes in environmental ecosystems. However, this integration is computationally challenging, as it requires solving repetitive LP problems at each time step and spatial grid, limiting the feasibility of FBA-based simulations in complex ecosystems. In this talk, we introduce an innovative simulation framework that overcomes these challenges by leveraging artificial neural networks (ANNs) as surrogate FBA models. These ANNs are trained on randomly sampled FBA solutions and incorporated into RTMs as algebraic source/sink terms. This approach significantly reduces computational costs while maintaining solution accuracy and robustness, enabling more efficient multi-dimensional ecosystem modeling. In the case study on Shewanella oneidensis MR-1, we demonstrate how our method effectively captures dynamic microbial metabolic processes, including metabolic shifts in response to nutrient depletion. Workshop participants will gain practical knowledge on implementing ANN-based surrogate models in RTMs and explore their potential applications in various microbial and environmental systems.