Parameter benchmarks

Parameters whose defaults were measured but did not need a study of their own. Everything here was run on 2026-06-20 unless a row says otherwise, on yeast-GEM (4,102 reactions), iJO1366 (2,583) and e_coli_core (95).

The parameters that did get a dedicated campaign are in the studies listed under Parameter tuning; the current value and one-line reason for every parameter on either side is in Tuned parameter defaults.

Two measurements that show MATLAB's default should change

fseof: MATLAB classifies targets with no tolerance at all

flux_eps is the floor below which a reaction's flux trend is treated as noise rather than signal. It does three jobs inside the classifier: skip a reaction whose flux is flat across the scan, skip one whose regression slope is indistinguishable from zero, and call a reaction a knockout when its flux at maximum enforcement is effectively zero.

On iJO1366:

flux_eps	Amplified	Knocked down or out
1e-8	18	414
1e-7	18	414
1e-6 (default)	18	393

The 21 extra targets admitted below 1e-6 all have a flux standard deviation around 5e-7 across the ten scan steps, Gurobi's primal feasibility tolerance accumulated across 2,583 reactions, which is floating-point summation noise rather than a metabolic trend. Reporting them would send a user to engineer reactions that are functionally zero throughout.

MATLAB's FSEOF.m compares each step against the previous one as bare floats, with no threshold of any kind. That is stricter than the 1e-8 row here, so 21 is a lower bound on what the MATLAB side would flag, not a measurement of it. Giving MATLAB a floor means adding one rather than exposing a value it already has.

remove_genes: the MATLAB default predicts growth after deleting an essential gene

blocked_reactions decides what happens to a reaction whose gene rule is emptied by a deletion. Deleting b1779 from e_coli_core removes the only gene for GAPD, which is essential for aerobic growth on glucose:

blocked_reactions	Reactions left	Predicted growth
'keep' (MATLAB)	95	0.874	GAPD survives with an empty rule and runs unconstrained
'remove' (Python)	94	0.000	GAPD is deleted, glycolysis breaks

Both behaviours are defensible, for different jobs. Keeping the reaction is right when editing gene annotations, where the network should not change as a result. Removing it is right for essentiality and engineering work, where a reaction with no enzyme should not carry flux. The defaults disagree about which job is the common one, and the MATLAB default silently produces a false negative in essentiality screens.

Literature and convention values, checked

These were swept and confirmed rather than changed. Both implementations agree on each, so there is nothing to reconcile; the value of the check is that the number is no longer taken on trust.

Parameter	Value	What the sweep showed
fseof.n_steps	10	Choi et al. 2010's value; 5 and 20 give the same target set on iJO1366
fseof.max_fraction	0.9	Choi et al. 2010; below 0.9 admits spurious targets, 0.99 finds nothing extra
fseof.correlation_threshold	0.9	Choi et al. 2010; 0.7 adds four spurious amplification targets
connect_blocked_reactions.eps	1.0	Correct for RAVEN-convention models, where exchanges carry ±1000 bounds
connect_blocked_reactions.allow_net_production	False	The stricter question, and the one the function exists to ask
fill_gaps_fast_lp.epsilon	1e-4	The fastGapFill paper's value
fill_gaps_milp.big_m	1000.0	Matches RAVEN-convention bounds
check_tasks.close_boundaries	False	Matches RAVEN's checkTasks, which published task lists were curated against; True closes the model's own exchanges so that the task is the whole boundary