Protein Structure Prediction

Protein Structure Prediction predicts a 3D structure from a protein sequence and can optionally include ligand or binding-related outputs when the selected model supports them.

Use it for

• predicting a structure from a protein FASTA;
• creating a structure file for the 3D viewer;
• testing protein-ligand settings with Boltz-2;
• building a sequence-to-structure pipeline.

Inputs

• Protein FASTA/FA/FAA/TXT file, or inline protein sequence.
• Optional ligand SMILES.
• Optional ligand CCD identifier.
• MSA server setting.
• Accelerator setting.
• Output format: mmCIF or PDB.
• Recycling steps and diffusion samples.

If both a file and inline sequence are provided, Liatir uses the file.

Compatible models

• Boltz-2 Local Structure & Binding

Chai-1 is implemented but hidden until Linux CUDA validation is available.

Outputs

• mmCIF or PDB structure file.
• JSON summary.
• Confidence scores when available.
• Affinity scores when available.
• Provenance.

How to read the result

The structure file is the main artifact. Open it in the structure viewer and check whether a file was actually produced.

Confidence and affinity outputs are model-derived estimates. They should guide inspection, not replace experimental validation.

CPU runs can be slow. For short tests, use one diffusion sample and modest recycling steps.

Technical details

Tool ID: ai-protein-structure

The direct runner and pipeline tool both create Jobs and Results records. A run that finishes without a structure file is treated as an error because the main scientific artifact is missing.