Verified, then logged.
Sun 22 Nov 2026 · 00:34 KST / 15:34 UTC한국어 | ENSearch /

Breaking02 / IT

NVIDIA describes RFD4-Proteina method for designing proteins

A 3D-printed ribbon model of a protein sits on a lab bench beside a multiwell plate of subtly tinted samples as a blurred gloved hand reaches in with a pipette. AI-generated image

NVIDIA introduced its RFD4-Proteina design method in a developer blog post on 21 Oct 2026. The post covers protein design using reward-guided inference-time search, a technique that steers a model's output with a scoring signal while the model generates it.

NVIDIA begins with the size of the design space. A sequence of eight amino acids, the building blocks of proteins, has more than 25 billion possible combinations, the company wrote. At 100 amino acids, the number of combinations exceeds the number of atoms in the observable universe.

NVIDIA wrote: "Protein design is a search problem." Its premise is that a protein that could lead to a new therapy lies somewhere in that space.

The title and introduction of the post give no release number or licence terms.