proteinmpnn

Boltz provides open-source biomolecular structure prediction using Boltz-1/Boltz-2 models, serving as an alternative to AlphaFold2. It specializes in predicting protein complexes, validating designed binders, and handling protein-ligand interactions. This skill is particularly useful when you need open-source structure prediction or want to leverage local GPU resources.

The alphafold skill uses AlphaFold2 to validate protein designs by predicting structures and calculating confidence metrics. It supports single-chain validation, binder-target complexes, and multi-chain predictions with AlphaFold-Multimer. For faster single-chain predictions, developers should use the esm skill instead.

BoltzGen is an all-atom diffusion model for protein design that generates both backbone and side-chain coordinates simultaneously. It is particularly suited for designing proteins around small molecules or ligands where precise binding geometries are required. Use this skill when you need side-chain-aware design from the start and are working with a YAML-based configuration.

BindCraft provides end-to-end protein binder design with joint backbone and sequence optimization and built-in AlphaFold2 validation. It's ideal for production-quality binder campaigns offering different speed protocols to balance design quality and computational cost. Use this skill when you need high experimental success rates for binder design rather than just backbone generation.