Empirical validation runs and parameter calibrations that back raven-python's defaults and its equivalence claims against MATLAB RAVEN.
- Human-GEM validation — raven-python ftINIT vs MATLAB RAVEN on 5 Hart2015 cell lines (Jaccard 0.975–0.980).
- (f)tINIT parameter calibration — clean-data calibration
plus input-robustness study (
mip_gap/big_m/force_on/eps/prod_weight/ scaling sweeps; dropout / noise / downsample robustness). - (f)tINIT solver benchmark — Gurobi vs HiGHS vs GLPK on genome-scale ftINIT.
- Yeast localization benchmark —
predict_localizationagainst curated yeast-GEM, with a predictor-noise sweep. - KEGG HMM cut-off calibration — HMM E-value / score-ratio sensitivity for the KEGG HMM-query reconstruction path.
:hidden:
humangem_validation
init_param_calibration
init_solver_benchmark
yeast_localization_benchmark
kegg_hmm_cutoff_calibration