In brief: A reproducible study needs far more than a screenshot of a promising generated sequence. Report enough detail for another researcher to understand what was done and what remains unverified.
Experiment identification
Record the research question, date, contact or project identifier, biosafety context and planned success criteria. State whether the result is exploratory, computationally evaluated or supported by laboratory experiments.
Data and software record
- Data source, retrieval date, license and filtering rules.
- Training/validation/test split logic and homology thresholds when relevant.
- Model family, checkpoint identifier and software dependency versions.
- Prompt inputs, track masks, sampling settings and random seed.
- Compute device, runtime, batch size and known failures.
Evaluation table to include
| Category | Report |
|---|---|
| Validity | Input and output validation checks |
| Baselines | Simple or previous models for comparison |
| Uncertainty | Replicates and confidence intervals |
| Failures | Excluded samples and why |
| Evidence | Computational vs experimentally measured |
Decision record
Write down what claim the results support, the leading alternative explanations and the next validation experiment. Explicitly identify gaps and avoid presenting simulated values as measured biochemical properties.
Primary sources and further reading
Editorial review: October 2026. This guide is educational. Model availability and APIs may change; confirm exact versions in the linked documentation. Generated structures and sequences require independent scientific validation.