RoseTTAFold is an open-source deep learning system for predicting protein structures, developed in the Baker lab ecosystem and maintained by RosettaCommons. With around 2,300 stars on GitHub, it was one of the first publicly available models to bring near-experimental-accuracy protein folding to any researcher with a GPU.
RoseTTAFold uses a three-track neural network that jointly reasons over a protein’s sequence, residue-to-residue distances and orientations, and 3D coordinates, refining them together across the network. Given an amino acid sequence (and an MSA built from sequence databases), it outputs a full 3D structure. The repository also covers protein-protein complex modeling, making it useful beyond single-chain prediction.
RoseTTAFold’s 2021 release, together with AlphaFold2, was recognized as Science magazine’s Breakthrough of the Year, because it made accurate structure prediction available to any lab rather than only to groups with specialized pipelines. Structures predicted with RoseTTAFold have since been used across biology — from understanding disease-related proteins to designing new protein binders — and the code remains a standard baseline that newer prediction methods compare themselves against.
Practically, using RoseTTAFold means preparing sequence databases and an MSA pipeline, and predictions run fastest on a CUDA GPU, though CPU runs are possible for small proteins. The repository’s issue tracker and the broader Rosetta community are active resources, and many later tools — including complex-prediction and design pipelines — reuse RoseTTAFold’s networks or its three-track design as components.
RoseTTAFold is free and open source under the MIT license. Running it requires your own compute and sequence databases.
Structural biologists, computational biochemists, and drug-discovery researchers who need open, self-hostable protein structure prediction.