[jplfaria]

Why this matters in our workflow

MSBuilder.build_metabolic_model cannot build any model on dev. The example notebook
examples/Model Reconstruction/build_metabolic_model.ipynb and the API docs (docs/source/core/msbuilder_api.rst)
both use it. We call build() and build_base_model directly, so our own builds are not affected. We found this
while reviewing dev's builder.

What happens

The builder refactor (f2de5f2, Sep 2025) renamed the old
build(model_or_id, index, allow_all_non_grp_reactions, annotate_with_rast, biomass_classic, biomass_gc, ...) to
build_base_model. It added a new build(model_or_id, gapfill_media, index, ...), which runs ATP correction and
gap-fill and returns a 4-tuple.
build_metabolic_model
still calls build() with the old positional arguments:

model = builder.build(model_id, index, allow_all_non_grp_reactions, annotate_with_rast, classic_biomass)

So every argument lands one slot off: index goes in as gapfill_media, allow_all_non_grp_reactions as index,
and so on. The base model then gets compartment index False (reactions named like rxn05294_cFalse), and
build_biomass raises TypeError: can only concatenate str (not "bool") to str. If that step passed, build() would
run ATP correction and gap-fill with "0" as the medium and return a tuple, which build_metabolic_model would then
pass to gapfill_model as a model.

How to reproduce

The reproduction is a three-gene genome built on the public GramNeg template (ModelSEEDTemplates v7.0):

import json
from modelseedpy import MSBuilder, MSGenome
from modelseedpy.core.msgenome import MSFeature
from modelseedpy.core.mstemplate import MSTemplateBuilder

template = MSTemplateBuilder.from_dict(json.load(open("GramNegModelTemplateV7.json"))).build()
genome = MSGenome()
features = []
for i, function in enumerate(["Pyruvate kinase (EC 2.7.1.40)", "Enolase (EC 4.2.1.11)",
                              "Phosphoglycerate kinase (EC 2.7.2.3)"]):
    f = MSFeature(f"g{i}", "M"); f.add_ontology_term("RAST", function); features.append(f)
genome.add_features(features)
MSBuilder.build_metabolic_model("toy", genome, template=template, annotate_with_rast=False, gapfill_model=False)

dev:

  File ".../modelseedpy/core/mstemplate.py", line 737, in build_biomass
    if "rxn13783_c" + index in model.reactions:
TypeError: can only concatenate str (not "bool") to str

On this branch, the call returns a model with 48 reactions.

The fix

The fix calls build_base_model with the same arguments. They are already in build_base_model's order, which is the
pre-refactor build() signature, so this restores the behaviour from before f2de5f2. gapfill_model then gap-fills
as before.

         builder = MSBuilder(genome, template)
-        model = builder.build(
+        model = builder.build_base_model(
             model_id,

Not changed here: build_from_annotaton_ontology. It has the same stale self.build(...) call
(L791),
but fixing that call alone would not make it work. Right after it:

• it calls rxn.genes(), but cobra's Reaction.genes is a frozenset, so this raises TypeError: 'frozenset' object is not callable;
• it calls anno_ont.get_feature(...), but AnnotationOntology has no such method;
• it indexes gene_term_hash with cobra genes, but the keys are annotation-ontology features.

That needs its own change and a decision on how the function should work.

No package code calls build_metabolic_model. The example notebook calls it with keywords, and they work after the fix.

Tests

tests/core/test_msbuilder.py::test_build_metabolic_model_builds_base_model monkeypatches build_base_model to
record its arguments. It checks that build_metabolic_model passes (model_id, index, allow_all_non_grp_reactions, annotate_with_rast, classic_biomass) and returns the model.

Against dev's code, the stub's empty model reaches add_atpm inside build():

E       KeyError: 'cpd00001_c0'
E           AttributeError: DictList has no attribute or entry cpd00001_c0
1 failed

With the fix, it passes. Full suite (Python 3.10, cobra 0.32.1, CPLEX 22.1.1): tests/core goes from 56 to 57
passed, and tests/ from 58 to 59. black 22.8.0 and the other pre-commit hooks pass.

Real-model check

We ran this on a genome-scale E. coli K-12 genome (RAST annotations, GramNeg template v7.1):

• With gapfill_model=False, the result (1,828 reactions) is identical to
  build_base_model("Keio", "0", False, False, False) in reaction ids, bounds, GPRs and equations.
• With the default gapfill_model=True on minimal glucose, dev raises the same TypeError. This branch returns
  1,834 reactions, and the model grows on glucose (0.7685). That growth comes from gapfill_model's simple gap-fill,
  without ATP tests, as before the refactor.

Related

• Issue: MSBuilder.build_metabolic_model and build_from_annotaton_ontology call build() with the old signature (and get_feature does not exist) jplfaria/ModelSEEDpy#11. This PR covers build_metabolic_model; the
  build_from_annotaton_ontology part stays open.

🤖 Generated with Claude Code