Incorporating Flux Sampling into a Minimal Assumption Dynamic Flux Balance Analysis Algorithm
Conference on Constraint-Based Reconstruction and Analysis (COBRA)
2018
5th Conference on Constraint-Based Reconstruction and Analysis (COBRA 2018)
General Submissions
Methods & Software 2
Tuesday, October 16, 2018 - 5:15pm to 5:40pm
Recent advances in sequencing technology have contributed to the rapid development of several highly predictive genome-scale models. Techniques like flux balance analysis (FBA) allow for relatively simple, mechanistically based, interrogation of these models. A drawback of FBA is the assumption of steady state. Dynamic flux balance analysis (DFBA) is an extension of this technique to non-steady state systems; a primary advantage of this approach is the low extra cost associated with developing dynamic models that are still mechanistically valid.
DFBA simulators typically solve successive FBA problems, and use the resultant fluxes to move the dynamic states of the system forward in time. A challenge associated with DFBA is the proper way of handling the degenerate solution space resulting from FBA. We propose a novel algorithm that takes advantage of the degenerate solutions of FBA to more closely match in vivo conditions. Indeed, it is well known that noisy gene expression results in variable enzyme concentrations between clonal cells; therefore, one expects the metabolic fluxes experienced by each cell to be different. Our algorithm models this behavior by sampling from the solution space, and uses these sampled fluxes to move the states of the system forward in aggregate. The resultant time varying flux distributions represent an extension of flux variability analysis to the dynamic setting. Our algorithm stands in contrast to the current state of the art simulators that utilize sequential linear optimization to find a unique flux for each state that is integrated. While this approach is mathematically robust, it requires further assumptions that might not be valid, or could be hard to justify. If the predicted transient flux distributions associated with our approach prove to accurately capture the in vivo behavior of an organism, it would allow experimentalists the ability to carefully interrogate proposed engineering strategies in the dynamic setting.
DFBA simulators typically solve successive FBA problems, and use the resultant fluxes to move the dynamic states of the system forward in time. A challenge associated with DFBA is the proper way of handling the degenerate solution space resulting from FBA. We propose a novel algorithm that takes advantage of the degenerate solutions of FBA to more closely match in vivo conditions. Indeed, it is well known that noisy gene expression results in variable enzyme concentrations between clonal cells; therefore, one expects the metabolic fluxes experienced by each cell to be different. Our algorithm models this behavior by sampling from the solution space, and uses these sampled fluxes to move the states of the system forward in aggregate. The resultant time varying flux distributions represent an extension of flux variability analysis to the dynamic setting. Our algorithm stands in contrast to the current state of the art simulators that utilize sequential linear optimization to find a unique flux for each state that is integrated. While this approach is mathematically robust, it requires further assumptions that might not be valid, or could be hard to justify. If the predicted transient flux distributions associated with our approach prove to accurately capture the in vivo behavior of an organism, it would allow experimentalists the ability to carefully interrogate proposed engineering strategies in the dynamic setting.