Open-Source Protein Design Model from Nvidia and the University of Washington: RFD4-Proteina

The University of Washington School of Medicine's Institute for Protein Design (IPD) and Nvidia have introduced a new generative artificial intelligence model named RFD4-Proteina, built on the BioNeMo AI platform. Combining biomolecular reasoning with search capabilities, this model aims to optimize the design process of novel proteins with specific functions.

According to data shared by the developers, RFD4-Proteina exhibits a 5 to 10 times higher success rate and faster production speed in tests such as protein binding, small molecule binding, enzyme scaffolding, and DNA binding compared to previous methods. Nobel laureate IPD Director David Baker stated that the model expands capacity in drug discovery and industrial processes such as the breakdown of pesticides or polymers.

Offered as open source under a BSD-3 license, the project is supported by organizations such as Amazon Bio Discovery, Anthropic, Google Cloud, and Hugging Face. The work was carried out with funding provided by the Howard Hughes Medical Institute and the Gates Foundation, led by UW PhD students Nathaniel Corley, Maxwell Kazman, Rafi Brent, and visiting student Kieran Didi.