Build a generative protein binder design pipeline
This BioNeMo blueprint is a full “binder design loop” for generating candidate protein/peptide binders to a target protein sequence. The workflow is designed to automate the typical in-silico pipeline: generate binder backbones, propose sequences, and then predict structures for both the binder and the binder–target complex so you can triage candidates before you ever touch a wet lab.
The reference implementation explicitly chains together well-known components: RFDiffusion to propose binder backbones, ProteinMPNN to generate sequences for those backbones, and AlphaFold2 / AlphaFold2-Multimer to predict structures and complexes. The outputs are predicted multimer structures in PDB format, which can then be scored or inspected to shortlist binders that look promising.
What to try first: run the notebook end-to-end on a small target sequence to understand the shape of the outputs and failure modes (e.g., unrealistic complexes or low-confidence structure predictions). Then experiment with how changing constraints (sequence length, scaffold choices, or selection thresholds) affects both runtime and candidate diversity. Be aware the blueprint is storage- and GPU-heavy; it’s intended for serious local infrastructure.
Source listing: https://build.nvidia.com/blueprints?filters=publisher%3Anvidia