name: dayhoff description: Dayhoff Atlas accelerates protein design by pairing trillion-token training data with hybrid Mamba-Transformer models for rapid sequence generation and mutation scoring. license: MIT
Activation Conditions
Activate when users ask for Dayhoff sequence generation, mutation scoring, dataset integration, prototype launch, or performance tuning.
Scope Boundaries
- Keep workflows aligned with responsible-use requirements from
docs/alignment-constitution.md. - Enforce safe execution patterns for backend/frontend startup.
Reference App Details
- Backend path:
assets/dayhoff-prototype/backend. - Frontend path:
assets/dayhoff-prototype/frontend. - Start backend in terminal A and frontend in terminal B.
- Avoid polling commands and warn about first-run model download time.
Workflow
- Confirm safety and credential constraints before sequence generation tasks.
- For prototype launch, run backend and frontend in separate terminals.
- Do not use polling commands to check service status.
- Start backend first, then frontend, and warn about first-run model download.
- Route feature and integration requests to the appropriate documentation.
Routing
docs/quick-start.mdfor hello world and launch steps.docs/application-patterns.mdfor generation and mutation-scoring methods.docs/data-integration.mdfor atlas dataset and file-format integration.docs/prototype-expansion.mdfor extending the reference app.docs/performance-guide.mdanddocs/troubleshooting.mdfor runtime optimization and debugging.
Domain Summary
Dayhoff Atlas combines large-scale metagenomic and synthetic protein corpora with hybrid model architectures to support rapid sequence ideation and mutation prioritization.
Reference Links
- Azure AI Foundry Model Catalog: https://ai.azure.com/catalog/models/Dayhoff-170m-GR
- GitHub: https://github.com/microsoft/dayhoff
- Research Paper: https://aka.ms/dayhoff/preprint
- Dayhoff Atlas Collection: https://huggingface.co/collections/microsoft/dayhoff-atlas-6866d679465a2685b06ee969