Reproduce “Protein Design with Guided Discrete Diffusion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nate Gruver, Samuel Stanton, Nathan C. Frey, Tim G. J. Rudner, Isidro Hotzel, Julien Lafrance-Vanasse, Arvind Rajpal, Kyunghyun Cho and 1 more
Researchers from NYU, Prescient Design, and Genentech introduce diffusioN Optimized Sampling (NOS), a method for gradient-guided discrete diffusion, integrated into LaMBO-2, a Bayesian optimization framework for protein design. LaMBO-2 generates enriched antibody libraries in silico, achieving 99% expression and 40% binding rates in preliminary in vitro validation, demonstrating efficient optimization with constrained edits.
Key insights
- Gradient-based optimization can be effectively applied to discrete diffusion models by leveraging gradients in the continuous hidden states of the denoising network.
- For constrained protein engineering (e.g., limited edits), intelligently selecting edit positions via saliency maps can be more impactful for objective improvement than the guidance mechanism itself.
- Iterative refinement (diffusive sampling) is crucial for significant objective improvement in protein design, allowing for modifications across the entire sequence rather than just autoregressive generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.