aniger-pilot — Aspergillus niger virtual microbe, unified pilot

Two A. niger models at different scales, plus the annotation crosswalk between them, in one clone-and-run repo. One verify.py checks all three layers (12 hard checks).

pip install -r requirements.txt
python verify.py   # exit 0 = all 12 checks passed

Why one repo with two models

Both models are Aspergillus niger, and they answer different questions:

Model	Scale	What it is	Growth
iJB1325 (BiGG)	2320 rxn / 1818 met / 1325 gene	published genome-scale model, A. niger ATCC 1015	0.9399 h⁻¹
aniger_ccm (in-house)	28 rxn / 24 met / 17 gene	curated central-carbon-metabolism core	18.95 model units (not h⁻¹)
crosswalk	—	reaction-level annotation mapping between them	—

They are not merged into one SBML file, and that is a measured decision (check G7 asserts it):

• metabolite ids: 24 vs 1818 → 0 collisions
• reaction ids: 28 vs 2320 → 0 collisions
• gene ids: 17 vs 1325 → 0 collisions — the CCM uses symbolic names (acoA, citA, cexA), iJB1325 uses numeric ids

More decisively: 13 of the CCM's 28 reactions have no genome-scale counterpart (EX_phos, PHOSt, EX_cit, EX_glucon, the lumped EMP_pyr/EMP_pep, ATP_sink, BIOMASS, DM_bio, BGC_NRPS, BGC_PKS, DM_sm1, DM_sm2). Merging would fabricate reactions that exist in neither model. Side by side, each model stays internally consistent and the mapping between them stays an explicit, inspectable artifact.

What the models actually show

iJB1325 — genome-scale, A. niger ATCC 1015. Reproduces BiGG's published dimensions exactly and grows at 0.9399 h⁻¹. Closing all 353 organic-carbon exchange boundaries collapses growth to 0 (100%), which proves flux accounting is carbon-bounded; it does not predict any phenotype.

aniger_ccm — a 28-reaction core with a real biomass reaction (BIOMASS, SBO:0000629, real precursors + GAM), genuine GPR, and a verifiable carbon guardrail (5 organic-carbon exchanges → 100% collapse).

Its headline result is the phosphate switch, which is the citric-acid production phenotype industrial strains are selected for:

Phase	Growth	Citrate secretion capacity
phosphate sufficient	18.95	6.00
phosphate depleted (EX_phos closed)	0.00	12.00 (2×)

One measured trap. You cannot read citrate overflow off a biomass-maximising FBA solution: with biomass as the sole objective the LP returns a carbon-minimal knife-edge solution and EX_cit is exactly 0.0000 in both phases — a false negative that hides the phenotype completely. Overflow is an alternative objective, so each phase is solved as "keep growth ≥ 50% of the phase-1 maximum, then maximise EX_cit".

Layout

models/genome_scale/iJB1325/iJB1325_ATCC1015.xml   BiGG full GEM (BiGG licence)
models/curated_ccm/aniger_ccm_refined.xml           in-house CCM (MIT)
src/aniger_fba.py                                   zero-dependency wrapper
crosswalk/aniger_annotation_crosswalk.{py,json}     mapping + its measured output
verify.py                                           12 hard checks, fail-closed
THIRD_PARTY_LICENSES/BiGG_LICENSE.txt               verbatim, must not be removed

Two rules this repo enforces in code

Both were learned by getting them wrong, not by reading about them.

1. Identity assertions never ride on the numeric checks. iJB1325 was originally published as ijb1325-ecoli-pilot, labelled E. coli. Every dimension and growth number matched published values and every check was green — while the organism label was wrong. It is A. niger ATCC 1015. Check G1 now asserts the organism of both models separately from any measurement, and G2 asserts the two growth units are not conflated.

2. Overflow products are invisible under the natural objective. See the measured trap above. Any downstream reuse of phosphate_switch() must keep the growth-floor structure.

Honest limits

• The CCM is a focused core, not whole-cell. Its lumped reactions (EMP_*, GOX, ATP_sink) do not strictly conserve mass/charge — a tool like MEMOTE would flag mass_balance on them. Deliberate simplification.
• Units are not comparable. 0.9399 h⁻¹ vs 18.95 model units.
• The crosswalk is partial and not validated as orthology. 15/28 reactions map (43 distinct iJB1325 targets), genes are propagated through GPRs with no sequence alignment, and one mapped target (BOUNDARY_bDGLCe, L-glucose) is a stereo mismatch recorded as a known defect (check G6c).
• No experimental validation. Matching BiGG's published dimensions proves the file is iJB1325, nothing more.

Licence

LICENSE (MIT) covers src/, verify.py and the curated CCM model. iJB1325 is NOT covered — BiGG models are not CC BY. Academic/non-profit use is free; commercial use requires contacting invent@ucsd.edu. THIRD_PARTY_LICENSES/BiGG_LICENSE.txt ships verbatim and must not be removed. See NOTICE.md.

Superseded

iMA871 (BioModels) is abandoned: gene=0, an artificial biomass sink, and a carbon guardrail that is not verifiable. The curated CCM supersedes it. EX_glucon in the CCM retains an iMA871-derived gluconate exchange route.

Series

field-claw/vitamind-virtual-microbe is the series hub.