Reviewers 1.2 / 1.3: transport scoring limits, and what template scope actually means

[freiburgermsu]

Answers reviewer 1's comments 2 (transport) and 3 (why only ~9,000 reactions reach a reconstruction). Text and docs only — no data or schema changes.

Every number below is measured on the live population (non-obsolete records) by Papers/NAR_Update_2026/analysis/review_transport_and_llm.py --live, added in the figures PR of this series, against the protonation-invariant tree (freiburgermsu#7, 1ad34adc). The dated sections at the end record how the numbers moved as the population and the tree changed.

---

Reviewer 1, comment 2 — transport

"Transport reactions do not have compartment information (they are only distinguished as compartments 0 and 1, unlike Rhea which identifies them as in and out). And the authors mention that transport reactions are scored from stoichiometry alone. I think it would be important to elaborate a bit more on this and discuss potential limitations (after all, thermodynamics are the most important for energy metabolism, which usually involves transport reactions)."

Where it came from. One sentence at the end of the conditions paragraph in M05_methods_thermodynamics.tex:

"A key caveat: as we publish a generalized database that can be applied across the kingdoms of life, transport reactions are scored from stoichiometry and energy alone, and neither an estimate of membrane potential or pH gradients are used."

It sat at the end of a paragraph about reporting conditions and was never picked up in Results or Discussion. The reviewer is right that this is under-treated.

What changed. The caveat is promoted to its own \paragraph{Transport.} and now makes three claims, each measured:

claim	measurement
the 0/1 convention is relative, not missing	documented in Biochemistry/REACTIONS.md, which previously called m only "a compartment index number"
an energy-only score cannot see the membrane	96.0% of the 1,186 gold-graded transport reactions carry ATP+ADP; only 2.7% translocate a proton
the scheme does not over-claim on trivial cases	of 2,688 transport reactions with no net chemistry, none is gold (1,715 bronze, 795 ungraded, 178 silver)

The first is the reviewer's factual point and it is a documentation gap, not a data gap: indices are relative so that one record can be instantiated against a bacterial plasma membrane, a mitochondrial inner membrane or a plastid envelope when a template builds a model. Rhea's in/out maps onto them directly.

The second is the substantive limitation, and it is sharper than "transport is approximate". A primary active transporter carries ATP hydrolysis in its stoichiometry; every predictor estimates that hydrolysis tightly; the reaction therefore earns a confident grade from chemistry that was never in question, while the translocation contributes nothing to the score. Transport is graded gold 34.4% of the time against 5.4% elsewhere, and that gap is almost entirely ATP coupling.

Also now recorded: dGPredictor commits to a direction on 1.4% of transport reactions against 17.3% elsewhere, so transport grades rest on fewer independent sources than anything else in the database — directly relevant to the reviewer's point that energy metabolism depends on these reactions.

Deliberately not done here. Flagging transport grades in the released records (an evidence_caveat key, or capping transport at silver) changes the shipped schema and is a call for the corresponding authors, not a reviewer-response edit. The paragraph is worded to describe only what ships today — is_transport, which is already in every record. Raised as a follow-up rather than assumed.

---

Reviewer 1, comment 3 — the ~9,000 reactions

"The authors mention that only 9.000 of the 56.000 reactions in the database are used for reconstructions. It would be great if they could elaborate on this. Is it because they are disconnected from the main core network? Is it due to lack of GPR associations?"

Where it came from. The opening sentence of M10_results_structure_curation.tex, which was scoping the curation pipeline — we prioritised curation on the reactions the templates touch — not asserting anything about usability. The reviewer read it as a coverage claim. Since a careful reader drew that conclusion, the sentence is rewritten rather than the reviewer corrected.

What changed. M10's opening now says plainly that the scope is a curation priority, and points at M13, where the question is answered.

Both of the reviewer's hypotheses are wrong, and the data says so:

• Not GPR. The Gram-negative v7.0 template types its 8,584 reactions explicitly: 6,284 conditional, 2,258 gapfilling, 31 spontaneous, 11 universal. A quarter of the template is present with no gene requirement at all. A template is not a set of reactions with genes attached, so missing GPRs cannot be the gate.
• Not connectivity. Templates are curated scopes; nothing prunes on graph connectivity.

The real limits, now stated in M13:

v7.0 template union	8,597 reactions
mass- and charge-balanced	33,496 (69%)
carry an EC number	20,016 (41%)
balanced and EC-annotated	15,128
…of those, outside every v7.0 template	8,826

That last row is the useful answer: roughly a doubling of template scope is already available, and the structure curation this paper reports is what makes it reachable. M14 gains one sentence naming compartment-aware scoring and the ~50% of reactions with no direction as the two standing limits.

---

Page budget

The NAR checklist in main.tex notes the draft is already at the limit. The two new paragraphs are offset by removing genuine duplication rather than by cutting content:

• M05's Uncertainty paragraph repeated the three median uncertainties (0.63 / 10.41 / 17.01 kcal mol⁻¹) verbatim from M11 Results. They now appear once, in Results.
• The Experimental anchors paragraph is tightened without losing a claim.
• A stray )) in the Figure 2C cross-reference is fixed in passing.

Net effect: body text still ends on page 6; six reference entries now spill to page 7 where the baseline fit in six pages. Three of the citations on that page (mavrovouniotis1993, xu2008, noor2014) are the orphaned ones addressed in the LLM PR of this series, so the spill is expected to resolve once that lands. \lastpage{6} should be re-checked after all four PRs merge.

Verification

• latexmk -pdf main.tex clean; no undefined references or citations.
• Supplement unchanged by this PR.

🤖 Generated with Claude Code

---

Merge note (added after the live-basis conversion)

This branch and #297 both touch M13_results_mapping.tex, which has a single-line body, so git cannot auto-merge them:

• Reviewers 1.2 / 1.3: transport scoring limits, and what template scope actually means #297 appends a \paragraph{Reconstruction scope.} after the atom-mapping sentence
• Reviewers 1.1 / 1.4: Euler for Fig 1B, a legend for Fig 2A, a pathway panel, and the growth basis #299 rewrites the atom-mapping sentence itself (stale numbers + live basis)

The resolution is to keep both, in this order:

Atom mappings are published for 26,242 reactions, 54\% of the database: 19,809 clean, and 6,433 that required salvage (which are tagged). As the mappings can be used to span a metabolic reconstruction, reactions needing one repair are likely to carry others.

\paragraph{Reconstruction scope.} A template carries far fewer reactions than the database holds ...

No other pair of branches in this series conflicts — verified with git merge-tree across all four.

---

Review pass (2026-09-23)

Self-review against the shipped data after the population decision in #299. One defect: the four percentages in the Transport paragraph were all-records values; they are now live — 96% ATP-coupled, 2.1% translocating a proton, dGPredictor 1.4% on transport vs 17.3% elsewhere. Re-verified unchanged: the template-scope numbers (8,597 / 2,258 gapfilling / 58% / 41% / 14,041 / 8,089) were computed on the live basis already; uniport-like reactions are graded gold on neither population (0 of 4,056 all-records, 0 of 2,688 live).

---

Re-measured against the protonation-invariant tree (2026-09-25)

#296 merged into dev on 2026-09-23. Its follow-up, freiburgermsu#7 (protonation-invariant-gate, 1ad34adc), fixes four defects in the structure pipeline and re-runs the reaction refresh, and it is the tree the paper now describes. Every number this branch introduced was re-measured there (commit e1c4f6b), and the tables above now show the right-hand column:

	on dev (41b20c2)	on #7 (1ad34ad)
gold-graded transport that translocates a proton (M05)	2.1%	2.7%
reactions satisfying mass and charge balance (M13)	28,096 (58%)	33,496 (69%)
balanced and EC-annotated	14,041	15,128
…of those, outside every v7.0 template	8,089	8,826

Unchanged: the template union (8,597), 2,258 gapfilling, 96% ATP-coupled, dGPredictor 1.4% / 17.3%, EC share 41%.

Why the balance share jumps eleven points. #7 repairs the reaction refresh itself: Rebuild_Stoichiometry.py had never refreshed the charge embedded in each reagent, and Rebalance_Reactions.py rejected its own save flag, so the proton-adjustment step had not been persisting. Between the two trees 5,428 live reactions go from imbalanced to balanced — 3,130 by a proton or water adjustment the fixed step now writes, 969 through a participant whose record the re-pick changed, and 1,329 whose flag had been stale against unchanged records and stoichiometry — and 25 go the other way, every one through a changed participant. The paragraph's claim, that balance is the first gate on template promotion, is unchanged; the gate is now measured on a refresh that actually ran.

Merge order. This branch is text-only against dev and merges cleanly onto #7 (git merge-tree, verified), but its numbers do not reproduce on dev alone: review_transport_and_llm.py --live run on this branch by itself returns the left column. Merge #7 first, then this. The M13 overlap with #299 is unchanged and resolves as the merge note above says, with the atom-mapping sentence at its re-measured values (26,242 / 19,809 / 6,433).