OpenFold3 on Amazon SageMaker JumpStart: Predict biomolecular structures
** OpenFold3‑preview (version 0.4.1, checkpoint of3‑p2‑155k) is now available as a SageMaker JumpStart model that can be deployed to an asynchronous inference endpoint on GPU‑enabled instances (ml.g6.4xlarge default). The service accepts protein, DNA, RNA sequences and SMILES strings, returning all‑atom structures in mmCIF with confidence metrics (pLDDT, PAE, PDE). Engineers can run multi‑GPU parallel predictions, supply custom MSAs, and autoscale the endpoint to zero when idle, turning a previously heavyweight workflow into a managed, cost‑effective AWS service. **KeyTakeaways:** - Model version 0.4.1, checkpoint of3‑p2‑155k delivers all‑atom predictions for proteins, nucleic acids, and small‑molecule ligands. - Deployed via SageMaker JumpStart to an asynchronous inference endpoint running on ml.g6.4xlarge (or ml.g6.2xlarge / ml.g6.12xlarge) instances.
⚡ Key Takeaways
- Model version 0.4.1, checkpoint of3‑p2‑155k delivers all‑atom predictions for proteins, nucleic acids, and small‑molecule ligands.
- Deployed via SageMaker JumpStart to an asynchronous inference endpoint running on ml.g6.4xlarge (or ml.g6.2xlarge / ml.g6.12xlarge) instances.
- Autoscaling with “scale‑to‑zero” reduces cost when the endpoint is idle, but GPU warm‑up still adds minutes‑scale latency per request.
- Integration uses the SageMaker asynchronous inference API; results are stored in Amazon S3 in mmCIF format with pLDDT/PAE/PDE scores.
- Requires a CUDA‑compatible PyTorch stack and large model checkpoints; the endpoint runs in a network‑isolated container to keep sequences private.
** By abstracting GPU provisioning, CUDA setup, and request queuing, SageMaker JumpStart lets drug‑discovery teams embed high‑accuracy structure prediction directly into production pipelines without managing complex infrastructure. **TechnicalLevel:** Advanced **TargetAudience:** ML Engineers, Computational Biologists **PracticalSteps:** - Deploy the OpenFold3‑preview model using the…
✅ Practical Steps
- Deploy the OpenFold3‑preview model using the SageMaker JumpStart console or SDK to create an asynchronous inference endpoint on a supported ml.g6 instance.
- Submit prediction jobs (protein/DNA/RNA sequences or SMILES ligands) to the endpoint and retrieve mmCIF results with confidence scores from the configured S3 bucket.
- Enable autoscaling with a scale‑to‑zero policy to automatically shut down GPU instances when no jobs are queued, minimizing idle costs.
Want the full story? Read the original article.
Read on AWS ML Blog ↗