Nature Biotechnology Response to Reviewers: Revision Guide

Do not answer each referee with an isolated experiment. A larger experiment package can still fail if technology identity, comparison, application, and claim refer to different versions of the contribution.

Technology papers often compare models, delivery systems, assays, editing strategies, manufacturing workflows, or computational tools that received different data, tuning, compute, reagents, operator expertise, or exclusion rules. Create one row per comparator with all information and intervention budgets visible.

For computational biotechnology, audit training data, leakage, split unit, preprocessing, labels, hyperparameter search, external data, compute, calibration, and failure slices. For wet-lab platforms, audit cargo, dose, formulation, lot, protocol, instrument, operator, timing, sample, detection threshold, and normalization.

The response should be calm and exact. If a referee could not see a distinction, first assume the manuscript did not make it inspectable.

In our pre-submission review work with Nature Biotechnology revisions, we inspect platform identity, predecessors, inputs, comparator budgets, code, reagents, protocols, delivery, dose, models, endpoints, safety, statistics, generalization, manufacturing, figures, source data, reporting, and claims. We audit each link between the evaluated platform and the released artifact, and we observe the patterns below when capability, benchmark, transfer, and translation evidence do not support the same claim. These are qualitative manuscript patterns, not confidential journal outcomes.

Pattern 1: the technology is the protagonist only in the title

The central biological result could have been produced by several existing methods, and the proposed technology is one enabling component. We identify the capability that is uniquely changed and test whether it controls the paper's strongest result. If not, we narrow the technology claim or reposition the study.

Pattern 2: comparator budgets are asymmetric

The proposed method uses curated data, updated reagents, repeated tuning, or expert rescue while published baselines use default settings or incomplete implementations. We align information and optimization budgets. Apparent superiority often narrows to one useful operating region.

Pattern 3: one convenient model becomes a general platform

A cell line, mouse model, donor set, production batch, or benchmark task supports broad claims. We choose an external test that challenges the most important dependency, not another replicate of the development environment.

Pattern 4: translation language hides a missing engineering step

Clinical, manufacturing, or deployment potential is emphasized while dose, delivery, stability, yield, cost, workflow, failure recovery, or material access remains unresolved. We separate proof of capability from readiness for translation and state the engineering work still required.

The distinctive information gain is platform alignment: technology identity, fair comparison, enabling performance, transfer, consequence, and reproducibility must support the same claim.

Pattern 5: the released artifact is not the evaluated artifact

The paper reports a model, protocol, vector, cell line, construct, or analysis pipeline that differs from the resource a reader can access. Versions, dependencies, preprocessing, reagent substitutions, or undocumented expert decisions explain part of the published result. In our Nature Biotechnology review work, we trace every figure to a versioned executable or physical artifact, reproduce a representative result from the release package, and list any element that cannot be shared. Reproducibility language is then limited to what an external group can actually inspect and run.

Check the response and revised platform evidence together.

One referee may ask for deeper molecular mechanism while another asks for cross-model validation, safety, or manufacturability. These are different uncertainties. Tell the editor which uncertainty controls the paper's central claim, what evidence was added for each, and which boundary remains.

Do not merge evidence volume into a single answer. A mechanistic perturbation does not establish clinical transport; a new model does not prove manufacturing robustness; and an accessible repository does not make an unfair benchmark fair.

Revision is not acceptance. The most serious rejection-on-revision risks are a technology that remains secondary to the biology, unfair comparisons, an external test that repeats the same dependency, unresolved off-target or safety evidence, inaccessible resources, and an abstract that keeps the original broad claim after mixed revision results.

Most failed rebuttals are formally complete but leave one controlling technology-level uncertainty unresolved. More experiments do not help when they answer easier questions.

Submit if: every comment is answered and located; technology identity is explicit; comparator budgets are matched; primary endpoints and failure cases are visible; external evidence challenges the key dependency; safety and translation language match the tests; and code, data, protocols, and materials are inspectable.

Think twice if: the platform still depends on one convenient model, negative results are excluded, safety is inferred from absence of a signal, baselines remain under-tuned, resources are promised later, or the abstract claims general use from a narrow operating range.

We reviewed current Nature Biotechnology submission and editorial-policy pages, current Nature Portfolio revision guidance, the PLOS response-writing canon, and Nature Computational Science response guidance. Current journal instructions and the decision letter control required files. The platform-to-consequence audit is Manusights analysis.

1. Put the editor's controlling issues before referee sections.
2. Answer every comment with action, result, claim impact, and location.
3. State why the biotechnology remains the paper's central advance.
4. Match data, tuning, compute, materials, dose, operator input, and endpoints across comparators.
5. Report negative, failed, and out-of-domain cases with prespecified exclusions.
6. Test the most important sample, model, site, task, manufacturing, or delivery boundary.
7. Separate platform capability from biological discovery and clinical readiness.
8. Reconcile safety, off-target, biodistribution, degradation, and failure claims.
9. Make code, data, protocols, materials, environments, and source evidence inspectable.
10. Synchronize response, manuscript, figures, supplement, repositories, abstract, and title.

Wait for 14 complete Search Console days before judging discovery for the exact Nature Biotechnology response query. On day 21, retain, revise, consolidate, or retire this owner using indexation, query match, impressions, clicks, and qualified revision-scan starts. The source cluster had no journal impressions and one preview start, so this page is an uncertain product-intent experiment rather than established search demand.

Nature Portfolio sources establish the public revision framework. The platform evidence chain is Manusights interpretation.