Amazon Web Services has made OpenFold3-preview available for deployment through Amazon SageMaker JumpStart, according to an AWS blog post. The OpenFold Consortium and the AlQuraishi Lab at Columbia University developed OpenFold3 as an open-source biomolecular structure prediction model.
Users can deploy OpenFold3 into their AWS accounts with a few lines of code to predict 3D structures of proteins, DNA, RNA, small-molecule ligands and the complexes they form. The endpoint runs with network isolation, which AWS says can help keep sequences within the user’s AWS account.
The service addresses infrastructure challenges that computational drug discovery teams face when running structure prediction models, which require GPU infrastructure, CUDA and PyTorch configurations, large model checkpoints and serving layers that handle requests taking minutes rather than milliseconds. OpenFold3 runs in an AWS-maintained container on a managed SageMaker endpoint designed to handle request queuing and GPU warmup, with optional autoscaling.
OpenFold3 is an all-atom structure prediction model that predicts joint 3D structures from molecular sequences. It can fold multi-chain protein complexes, protein–nucleic acid assemblies and protein–ligand complexes in single predictions. The model accepts protein, DNA and RNA sequences, plus small-molecule ligands as SMILES, and outputs predicted structures in mmCIF format with confidence metrics including pLDDT, PAE and PDE.
The deployment uses OpenFold3 version 0.4.1 with the of3-p2-155k checkpoint, released under an Apache 2.0 license. Supported instance types include ml.g6.2xlarge, ml.g6.4xlarge (default) and ml.g6.12xlarge for running multiple predictions in parallel on multi-GPU instances. Users can supply their own multiple sequence alignments to improve prediction accuracy while keeping sequences private, and can scale endpoints to zero when idle to reduce costs.