AI Fold Floes Release Notes

v0.2.7 August 2026

This package is built using OpenEye-orionplatform==6.8.3, OpenEye-toolkits==2026.1.0, and OpenEye-Snowball==0.32.0.

The output mmCIF file generated by the Boltz prediction cube and saved to the output record now contains _chem_comp_atom and _chem_comp_bond fields for residues and ligands.
The Protein Sequence to AI Folded Structure Prediction Floe now strictly enforces the input ligand SMILES during Spruce preparation.
The Protein Sequence to AI Folded Structure Prediction Floe now runs pose checking to compare the Boltz predicted structure to the input ligand SMILES. If a ligand fails a pose check the output structure will need to be reviewed.
Better optimization has been added for CCD-coded ligand inputs during Spruce preparation.

This package is built using OpenEye-orionplatform==6.8.3, OpenEye-toolkits==2025.2.2, and OpenEye-Snowball==0.31.4.

Documentation has been added to guide the user on the Protein Sequence to AI Folded Structure Ligand Affinities Floe.
Documentation has been added to guide the user to create their own distributed MSA databases on Orion.
Documentation has been added to guide the user on different MSA databases that can be used with protein folding.

Support has been added to supply template structures to Boltz to guide the folding process.
General error handling has been updated.
Minor typos have been fixed.
Input form examples are no longer quoted to avoid confusion with the actual input forms.
Output from the Protein Sequence to AI Folded Structure Ligand Affinities Floe now have unique names using a count incrementing method.v0.2.6 April 2026

This package is built using OpenEye-orionplatform==6.5.0, OpenEye-toolkits==2024.2.1, and OpenEye-Snowball==0.29.2.

The Protein Sequence to AI Folded Structure Prediction Floe has been updated to run using the Boltz model. An optional MSA search prior to structure prediction has been added.
The Protein Sequence to AI Folded Structure Ligand Affinities Floe has been added, which takes an ensemble of input ligands that will be individually co-folded with the input sequences. Ligands with a higher affinity values will be ranked higher.
The MSA Align and Search Floe has been added, which runs an MMSeqs2 MSA search on input sequences.
The MSA Collection Setup from FASTA Floe has been added, which creates a FASTA collection that can be used with floes that perform an MSA search.

This package is built using OpenEye-orionplatform==6.0.0, OpenEye-toolkits==2023.2.3, and OpenEye-Snowball==0.27.1.

The Protein Sequence to AI Folded Structure Prediction Floe has been updated to run minimization and can optionally run pocket finding on the generated structure.

This is the first release of the OpenEye AI Fold Floes.
This package is built using OpenEye-orionplatform==6.0.0, OpenEye-toolkits==2023.2.3, and OpenEye-Snowball==0.27.0.

This package contains the following Floe, whose purpose is to generate 3D protein structures from only a sequence input.

Protein Sequence to AI Folded Structure Prediction: A protein sequence is used as an input for AI folding methods to predict protein structure. Currently, OmegaFold is the supported AI model to fold proteins.