Updated for 2026: Reproducibility checklist for ESM3 studies: model provenance, family-aware splits, complete metadata and validation.
Freeze the experimental context
Record the model checkpoint and access route, package versions, software dependencies, hardware, generation settings and input provenance. Keep a frozen requirements file or lockfile as part of your methods.
Design honest splits
Random splits of highly homologous protein sequences can leak similar examples between training and evaluation. Consider sequence-identity and protein-family separation where justified, and document deduplication methods.
Report uncertainty and baselines
Provide appropriate baseline results, failure cases, input distributions and uncertainty. If you use computational predictions as proxies for activity or stability, label them as proxies rather than measured biological performance.
Share carefully
Publish reproducible scripts, safe illustrative data and permissioned artifacts where possible. Respect licenses, privacy, security constraints and any requirements for independent experimental validation.
Evidence and further reading
Primary sources: Biohub ESM3 model, Biohub maintained ESM repository, and Biohub learning resources. Check model cards and release notes for version-specific instructions.
Editorial note: Updated October 2026. Distinguish scientific literature and current product documentation from speculative applications; verify current API details before running code.