Protein Design with RFD4-Proteina and Hugging Face Spaces
huggingface ·
NVIDIA and the University of Washington have co-developed RFD4-Proteina, a BioNeMo open model for programmable protein design that integrates sequence and structure generation into a single workflow. By combining a flow-matching model with an autoencoder, the system reduces information loss typically caused by handoffs between separate models. Users can access the tool via Hugging Face Spaces, with NVIDIA sponsoring free GPU access from October 21 to November 4, 2026. The model features a 'thinking setting' that allows for varying levels of computational refinement, significantly improving design efficiency compared to previous methods. Detailed benchmarks indicate that high-thinking settings produce substantially more structurally distinct designs per GPU-hour than competing approaches. The article provides comprehensive guidance on running designs through command line, Python scripts, or the Diffusers library, including specific code examples for PD-L1 binder generation. Additionally, the platform includes SecureDNA biosecurity screening to ensure compliance while allowing users to participate in external protein design competitions.
huggingface