Nature Genetics' AI Policy: Springer Nature Rules for Genomics and Human Genetics Authors

While the policy text is identical to Nature's, several aspects of genetics research create practical complications:

Genetic data privacy

This is the biggest concern specific to genetics journals. Individual-level genetic data, genotypes, sequences, variant calls, is among the most sensitive data in biomedical research. Most genetic datasets are governed by data use agreements (DUAs) that restrict how the data can be processed and where it can be stored.

If you input genetic data into ChatGPT, Claude, or any other cloud-based AI tool, you've almost certainly violated your DUA. These tools send data to external servers, often with terms of service that allow the provider to use inputs for model training. This creates three problems:

1. Privacy violation. Individual genetic data can identify participants, even when "de-identified." Feeding it to a cloud AI tool breaches most institutional data governance policies.
2. DUA breach. Datasets from dbGaP, UK Biobank, and other repositories have explicit restrictions on data processing environments. Cloud AI tools don't qualify.
3. Regulatory exposure. Depending on jurisdiction, genetic data may be covered by GDPR, GINA (in the US), or equivalent laws that restrict how it can be processed.

Nature Genetics' AI policy doesn't explicitly address genetic data privacy in AI prompts, but the journal's broader research ethics requirements do. Don't let an AI disclosure issue turn into a data governance crisis.

Computational analysis is everywhere

A typical Nature Genetics paper includes dozens of computational steps: variant calling, quality control, imputation, association testing, fine-mapping, functional annotation, colocalization, Mendelian randomization. Many researchers now use AI to help write or debug the code for these pipelines.

This is permitted and should be disclosed. But it needs careful framing:

Good approach: Describe AI-assisted code generation in your Methods as part of your computational pipeline disclosure. Confirm that the code was validated independently.

Bad approach: Lumping code assistance with language editing in a single vague statement. Reviewers at Nature Genetics will want to know specifically which analysis code was AI-assisted because it affects their assessment of the pipeline's reliability.

Variant interpretation

AI tools are increasingly used for variant interpretation, classifying variants as pathogenic, likely pathogenic, VUS, etc. If your paper includes variant interpretation that used AI-based tools (like AlphaMissense, CADD, or REVEL), these should be described as analysis methods in your standard Methods section, not in the AI manuscript preparation disclosure.

However, if you used ChatGPT to help write the text describing your variant interpretation results, that's a manuscript preparation issue and requires disclosure. The distinction: the tool that scored the variant is a research method; the tool that helped you write about the score is a writing aid.

For a GWAS paper:

"During the preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI) to improve the language clarity of the Discussion section. GitHub Copilot (Microsoft) was used to assist with writing quality control scripts for the genotype data processing pipeline; all scripts were validated against established QC protocols and manually checked by the bioinformatics team (L.M. and P.R.). No individual-level genetic data was processed through any cloud-based AI tool. The authors take full responsibility for the published content."

For a clinical genetics paper:

"The authors used Claude (Claude 3.5, Anthropic) to improve the readability of the Introduction and Methods sections. No AI tools were used for variant classification, clinical interpretation, or patient phenotyping. All genetic counseling implications described in the manuscript reflect the clinical genetics team's assessment. The authors take full responsibility for the content."

For a functional genomics paper:

"During manuscript preparation, the authors used ChatGPT (GPT-4, OpenAI) to assist with editing the Results section for conciseness. The single-cell RNA-seq analysis pipeline (Seurat, Scanpy) and CRISPR screen analysis code were written by the bioinformatics team without AI assistance. The authors take full responsibility for the published content."

What to avoid

"AI tools were used for data analysis and manuscript preparation."

At Nature Genetics, this is dangerously vague. "Data analysis" could mean anything from quality control scripts to variant interpretation to GWAS association testing. Reviewers will demand specifics, and the ambiguity may raise concerns about the analysis pipeline's integrity.

The imputation example is instructive. If you used IMPUTE2 or Michigan Imputation Server, that's a standard analysis tool described in Methods. If you used ChatGPT to help write a custom imputation wrapper script, that's AI-assisted code requiring disclosure. The imputation software itself isn't covered by the AI policy; the writing of scripts around it might be.

Standard Springer Nature enforcement:

During review: Editor requests disclosure. Nature Genetics' reviewers are computational experts who may flag AI-generated code patterns or text.

After publication:

• Correction for minor language editing non-disclosure
• Expression of concern if AI affected analysis code or interpretation
• Retraction if AI generated fabricated results or false genetic claims

Genetics-specific escalation risks:

If your paper involves human genetic data and undisclosed AI use raises concerns about data handling, the consequences can extend beyond the journal:

• Your institution's IRB or ethics committee may investigate
• The data repository (dbGaP, UK Biobank) may audit your data access
• Funding agencies (NIH, Wellcome Trust) may require an explanation
• Collaborating institutions may review their data sharing agreements with you

This isn't about language editing. If someone suspects that individual-level genetic data was processed through a cloud AI tool, the investigation becomes a data governance issue, not just a publication ethics question. The stakes are categorically different from what you'd face at a journal that doesn't routinely handle sensitive human data.

American Journal of Human Genetics (AJHG) uses STAR Methods because it's a Cell Press journal. Genetics in Medicine (GIM) focuses on clinical genetics and has the highest sensitivity to AI use in variant interpretation and genetic counseling language. Nature Genetics falls in between, it publishes both population genetics and clinical genetics, requiring awareness of both computational and clinical AI considerations.

The consortia factor deserves emphasis. Nature Genetics papers frequently involve 50-200+ authors across dozens of institutions. Ensuring that AI disclosure is consistent across all contributing sites is a genuine coordination challenge. If one site used AI to write their methods contribution and didn't tell the corresponding author, the paper's disclosure is incomplete. Large genetics consortia need AI use policies that extend to all participating centers.

For GWAS and population genetics:

• Never input individual-level genetic data into cloud AI tools. Even summary statistics should be handled carefully if they could be used for re-identification.
• If AI helped with analysis code, deposit the code in a public repository and note which portions were AI-assisted.
• For consortia papers, circulate an AI use survey to all contributing sites before submission.

For clinical genetics and variant interpretation:

• Don't use AI to generate variant classification language. ACMG criteria application should be done by trained clinical geneticists.
• If you used AI-based variant prediction tools (CADD, REVEL, AlphaMissense), describe them as research methods, not in the AI manuscript preparation disclosure.

For functional genomics:

• If AI helped with CRISPR guide design code, disclose this and confirm the guides were independently validated.
• For single-cell analysis pipelines, specify which steps used AI-assisted code and which used established packages.

Before submission checklist:

• [ ] AI disclosure in Methods with tool names, versions, and use cases
• [ ] No individual-level genetic data processed through cloud AI tools
• [ ] Research AI tools (CADD, AlphaMissense, etc.) described in standard Methods
• [ ] Manuscript preparation AI tools described in separate disclosure
• [ ] All analysis code deposited and documented
• [ ] AI-assisted code validated independently
• [ ] All co-authors (including consortium members) aware of AI disclosure
• [ ] Graphical abstract created with standard tools (not generative AI)

A free manuscript assessment can help you verify that your Nature Genetics submission meets the journal's editorial and ethical requirements before you submit.