Midnighter
commented
9 months ago Hey Ronan,
Thank you for the comment. Am I correct in interpreting your first paragraph to also say that if the heuristics succeed in detecting all artificial, inconsistent reactions, then the stoichiometric consistency detection is actually correct, though?
I must admit that it has been more than a decade for me that I last read some Matlab code. So to put it carefully, the function that you link to is hard to read. I will go through your paper and see if I can understand that better.
rmtfleming
The current implementation of stoichiometric consistency https://github.com/opencobra/memote/blob/master/src/memote/support/consistency.py#L66 provides a substantial under-estimate of the stoichiometric consistency when the heuristic check for internal reactions https://github.com/opencobra/memote/blob/master/src/memote/support/consistency.py#L93 misses some reactions that are actually external reactions and therefore, by definition, stoichiometrically inconsistent.
This COBRA Toolbox function uses a sequence of LPs that iteratively removes internal reactions of unknown, but probable stoichiometric inconsistency, e.g., reactions with large numbers of non-zero stoichiometric coefficients, non-integer coefficients, etc. https://github.com/opencobra/cobratoolbox/blob/master/src/reconstruction/modelGeneration/stoichConsistency/findStoichConsistentSubset.m
The algorithm underlying the aforementioned function and several others for checking model consistency and cardinality optimisation in general are described here:
Cardinality optimization in constraint-based modelling: application to human metabolism Ronan M T Fleming, Hulda S Haraldsdottir, Le Hoai Minh, Phan Tu Vuong, Thomas Hankemeier, Ines Thiele Bioinformatics, Volume 39, Issue 9, September 2023, btad450, https://doi.org/10.1093/bioinformatics/btad450