Comparing Flux Sampling Methods for Genome-Scale Microbial Models
Krista Williston
Mentor: Dr. Ali Navid, Biosciences and Biotechnology Division, Lawrence Livermore National Laboratory.
Date/Time: August 21st, 2025 at 1:15 PM.
Abstract: Computational modeling of microbial metabolism is critical for analyzing omics data and understanding cellular behavior. System-level modeling has applications in a variety of scientific and industrial areas such as metabolic engineering and drug discovery. Flux Balance Analysis (FBA) is a commonly used method for this purpose. FBA uses information available from annotated genomes to mathematically formulate the entire metabolism of a biological entity. The models are constrained using experimental data and then linear programming is used to optimize a single biological objective, such as growth. While this assumption might be reasonable for some types of analyses, it usually does not reflect real biological conditions. The uniform flux sampling analysis (FSA) is an alternate method for examining phenotypic behaviors of a system. Several different methods can be used for FSA. However, different FSA algorithms behave differently and may not always uniformly sample the many dimensional solution space of the models and might miss the most relevant biological solutions.
In this project, I evaluated several FSA algorithms, including Critical Hit-and-Run with Rounding (CHRR), Hit-and-Run (HR), Artificial Centering Hit-and-Run (ACHR), and Adaptive Direction Sampling on a Box (ADSB). For my initial tests, I used two sub genome-scale models, an E. coli core metabolism model and a model of metabolism in algal chloroplast. I then extended the analysis to a genome-scale E. coli model. The result of my initial analyses showed that for all the examined methods, the predicted flux patterns clustered near the center of the models’ solution space. In order to fix this shortcoming, I developed a method to use the observed characteristic of the methods to sample the entire solution space. My method involves changing the constraints of the models to adjust the lower biomass bounds and force sampling away from the center of the solution space. By systematically adjusting biomass constraints, I was able to improve sampling coverage and obtain more uniform flux distributions. Overall, this work demonstrates the limitations of common FSA algorithms and presents an approach to better explore the entire solution space of constraint-based modeling. These results will be more representative of all phenotypes throughout the system and provide a foundation for applying improved sampling methods to larger microbial models and eventually to more complex microbial communities