Design Proteins with RFD4-Proteina Using Reward-Guided Inference-Time Search
Protein design is a search problem. Choose amino acids for an eight-residue protein sequence and more than 25 billion combinations are possible. Choose amino...
First seen by Oriel on .
AI summary
The tutorial demonstrates using RFD4-Proteina to generate protein binders and steer candidate generation with inference-time rewards, including an example targeting PD-L1.
- RFD4-Proteina jointly generates protein sequences and all-atom structures, using a residue-level latent representation and an autoencoder decoder rather than a separate inverse-folding model.
- Its search loop scores denoising trajectories, retains and branches promising partial designs, and can use geometry, interaction, or prediction-confidence rewards.
- The PD-L1 example fixes the target structure and sequence, permits side-chain repacking, and designs a 100–150-residue binder around hotspots on Tyr56, Met115, and Tyr123.
- Open-model and NIM workflows share condition specifications; the tutorial provides generation and OpenFold3-guided search examples with two branches and beam width three.
- OpenFold3 pLDDT measures prediction confidence, not binding evidence; the tutorial recommends separately checking interfaces, refolding, clashes, constraints, and structural diversity before experimental testing.
#protein design #rfd4-proteina #inference-time search #drug discovery
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| NVIDIA Developer Blog | https://developer.nvidia.com/blog/?p=123747 | https://developer.nvidia.com/blog/design-proteins-with-rfd4-proteina-using-reward-guided-inference-time-search/ Read discussion |