Activity in progress: Batch Structure Prediction Workflows

Details

• Structure prediction software often natively supports single predictions.
• Specific workflows can re-use MSA input to massively improve efficiency over naive implementation.
  • Interaction screening where 1 protein is predicted in combination with many others.
  • Conformation sampling where 1 system is predicted with downsampled MSA with many random seeds.
  • Epitope mapping where 1 antibody-antigen system is predicted with many seeds to identify candidate binding epitope.
  • Stoichiometry screen where 1 system is predicted with many candidate stoichiometries.

This activity involves developing optimized batch workflows for structure prediction.

Completed

• Add support for re-using MSAs to Galaxy Australia AlphaFold2 service.

In Progress

• Add local implementation of batch colabfoldsearch to the nfcore proteinfold pipeline.
• Add mmseqs-GPU support to the nfcore proteinfold pipeline.
• Add boltz interaction screening workflow to screen multiple potential partners against a protein (re-using MSA for anchor protein).
• Add option for extreme replicate sampling for AlphaFold2.

Future

• Develop Galaxy workflow to take advantage of re-using MSAs.
• Add stoichiometry screening workflow to the nfcore proteinfold pipeline.

Contributors