MSA

Python API · stjames models · Boltz single MSA example · Boltz paired MSA example

How it works

Rowan's multiple sequence alignment (MSA) workflow searches for sequences related to your protein chains and aligns them for use in structure prediction. These alignments provide evolutionary information to models such as Boltz and Chai-1. The workflow generates alignment files; structure prediction is a separate calculation.

Settings

• Protein sequences: provide one or more protein chains through the web interface or Python API. For a protein complex, include its chains together.
• Output formats: select one or more of "ColabFold," "Boltz," and "Chai" under "Output formats." The default is "ColabFold." Choose the format expected by your downstream prediction model.
• Pairing: unpaired alignments are always generated. When you supply multiple sequences, the workflow also generates paired alignments automatically; there is no separate pairing setting.

Notes

The search uses a Rowan-hosted ColabFold/MMseqs2 service at msa.rowansci.com, with UniRef and environmental sequence searches. Unpaired alignments contain related sequences for each chain independently. Paired alignments associate related sequences across chains to help a model infer contacts within a protein complex. A paired alignment is useful input for complex prediction, but it does not establish that the submitted proteins interact.

Output formats

Each selected format produces its own compressed archive. After completion, select a format and use "Download MSA" to retrieve it, or download the files through the Python API.

Boltz and Chai files combine paired and unpaired alignment information. Keep the original protein sequences with the downloaded files so the alignments can be matched to the correct chains when used in a later prediction.

Further reading

• MSA, protein–ligand binding affinity exploration, and stereochemistry
• Using securely generated MSAs to run Boltz-2 and Chai-1