Question about bounding the model and downstream analysis with ftINIT algorithm

[Edward-NiAn]

Hello,

I have some questions regarding the utilization of the ftINIT algorithm with GEMs other than the human model.

I adapted the full workflow for a different GEM and scSeq data by replacing the corresponding metabolites in the essential metabolic task file provided from the tutorial. The grouping of generated models was successful and met my expectations. However, I encountered issues with downstream FBA. Despite the models having only about 100 different reactions, the fluxes varied significantly. When I applied nutrient boundaries based on experimental data, most of the generated models returned infeasible solutions with solveLP(), while the same boundaries would work with the reference model.

My questions are:

1. Is the variation in FBA results inevitable due to model reduction? Are there any existing solutions or relevant papers addressing this issue?
2. Would it be better to impose nutrient boundaries before running the ftINIT step? My preliminary attempts suggest that ftINIT often does not return satisfactory solutions when nutrient boundaries are imposed initially, but I haven't explored this extensively.

Any help or advice would be greatly appreciated. Thank you!