boltzina

A virtual screening pipeline that runs Boltz-2 structure and affinity prediction, then docks with AutoDock Vina or Uni-Dock2 and scores affinity. Input can be a sequence or a Boltz YAML.

Boltzina combines AutoDock Vina or Uni-Dock2 docking with Boltz-2 structure prediction and affinity scoring, packaged as an efficient and accurate virtual screening approach built on docking guided binding prediction. The docking and the prediction are separate stages that feed into one another. The two docking backends are offered as alternatives, so users pick based on what they already use.

The sequence mode takes a protein sequence and a SMILES or SDF file. Boltzina runs Boltz-2 structure and affinity prediction for the complex with the first, or reference, ligand, then continues into the next pipeline steps from there.

For multi chain proteins, ligand definitions, and Boltz-2 settings, a boltz compatible YAML file gives full control, with an example provided. More protein chains simply mean more protein entries, and the affinity binder property identifies the reference ligand used for grid center determination.

Run options include a protein sequence flag that accepts single or multi chain sequences, a FASTA flag with one entry per chain, and a YAML flag for a Boltz compatible configuration. The reference ligand defaults to the first in the input, the grid center defaults to automatic from the predicted binding pose, and the grid size defaults to 20 angstroms.

Given precomputed Boltz-2 results, the grid center is taken automatically from the predicted ligand position and can be overridden explicitly. A ligand chain id option supports rescore mode, which rounds out the documented configuration surface.

• Source 1 · Project repository
• Source 2 · README