NVIDIA Unveils Open Protein Design Model as AI Expands Into Drug Discovery and Cancer Care
The new RFD4-Proteina model, developed with the University of Washington’s Institute for Protein Design, leads a wider push to connect AI-driven biology, autonomous laboratories and more personalized cancer treatment.
NVIDIA is expanding its role in healthcare and life sciences with a new open protein design model and collaborations aimed at accelerating drug discovery, automating scientific experiments and improving cancer treatment planning.
Announced at GTC Berlin, the initiatives bring together the University of Washington’s Institute for Protein Design, Siemens Healthineers and a growing group of biotechnology, laboratory technology and healthcare companies. The projects span computational protein engineering, AI-powered scientific agents and adaptive radiotherapy, where treatment plans can be adjusted to individual patients.
The centerpiece is RFD4-Proteina, a new model on NVIDIA’s BioNeMo platform that combines programmable protein design with search capabilities. NVIDIA says the model is designed to help researchers generate potential protein candidates, evaluate them against specific objectives and prioritize which designs should move into laboratory testing.
The broader announcement also highlights a shift toward AI systems that do more than generate predictions. Scientific agents are being developed to connect research data, specialized models and laboratory equipment, potentially helping scientists move from a hypothesis to an experiment with fewer manual steps.
“Agentic and physical AI are opening new possibilities across medicine, from designing proteins and powering autonomous laboratories to improving personalized cancer care,” said Kimberly Powell, NVIDIA’s vice president of healthcare.
RFD4-Proteina Aims to Make Protein Design More Targeted
Proteins carry out essential biological functions, from enabling chemical reactions to helping cells recognize and respond to other molecules. Designing new proteins with specific properties could support future medicines, diagnostic tests and industrial applications.
The difficulty is the sheer number of possible molecular structures. Researchers must identify designs that are not only computationally promising but also capable of performing their intended function under real experimental conditions.
RFD4-Proteina is intended to make that search more efficient. Scientists can provide molecular inputs, structural constraints and scoring objectives, allowing the model to generate and refine candidate proteins rather than simply producing designs without a targeted evaluation process.
According to NVIDIA’s announcement, a research draft reported improvements in computational design success rates and generation throughput over the evaluated baselines across four benchmark areas: protein binding, small-molecule binding, enzyme scaffolding and DNA binding. These are computational results, however, and should not be confused with proof that a candidate will work as a medicine or diagnostic tool.
The model’s potential applications include designing proteins that bind to disease-related targets, support diagnostic detection or perform specialized chemical functions. Each candidate still needs appropriate laboratory evaluation, and therapeutic applications would require further development and testing.
David Baker, the 2024 Nobel laureate in chemistry and director of the University of Washington’s Institute for Protein Design, said the model could expand the range of proteins researchers are able to create.
“Making the model open will allow researchers around the world to build on these capabilities and pursue solutions once beyond the reach,” Baker said.
The release is also designed to make the technology accessible through existing research and computing environments. NVIDIA lists Amazon Bio Discovery, Anthropic’s Claude Science, Benchling, Databricks, Google Cloud, Hugging Face, Nebius, OpenAI and Tamarind Bio among the partners through which researchers can access or integrate RFD4-Proteina.
Siemens Healthineers and NVIDIA Target More Adaptive Cancer Treatment
The second major initiative focuses on a different problem: helping clinicians develop precise cancer treatment plans while managing complex imaging data, dose calculations and time-sensitive decisions.
Siemens Healthineers and NVIDIA are collaborating on the next generation of adaptive radiotherapy, combining AI models, accelerated computing, patient-specific simulation and real-time data processing. The goal is to help care teams evaluate treatment options and adjust them to individual patient needs.
Adaptive radiotherapy can account for changes in a patient’s anatomy or tumor during a course of treatment. The underlying challenge is to maintain the prescribed treatment effect on the tumor while limiting radiation exposure to nearby healthy tissue.
The proposed NVIDIA-enabled workflow has several components:
- AI-assisted treatment planning: NVIDIA Nemotron open foundation models are being explored for agentic co-planning, helping clinicians assess and refine personalized treatment strategies.
- Patient-specific simulations: CUDA-accelerated dose calculations could help teams evaluate radiation delivery more quickly for individual patients.
- Real-time data integration: NVIDIA Holoscan could connect sensor data with AI processing and simulation to support adaptive treatment and robotic control.
These technologies are being developed as part of a collaboration; the announcement does not establish that every component is already deployed as a routine clinical system.
Arthur Kaindl, head of Varian, Siemens Healthineers’ cancer-care tech-ai-spending-2026/">business, said the partnership builds on existing progress in treatment precision and adaptation, including the company’s Accela system.
The companies are exploring how AI and real-time computing could give clinicians greater speed and insight when making treatment decisions. Any improvement in patient outcomes, treatment access or clinical efficiency will depend on the performance and validation of the resulting systems.
Scientific Agents Could Connect Drug Design With Laboratory Experiments
NVIDIA is also expanding the BioNeMo ecosystem, which provides an open development platform for AI-driven biology and drug discovery. The objective is to connect specialized scientific models with research data, reasoning systems and laboratory workflows.
In conventional research pipelines, computational predictions and physical experiments can involve separate tools and repeated manual handoffs. Scientific agents could help coordinate parts of that process: selecting an appropriate model, evaluating its output, organizing an experiment and using the resulting data to inform subsequent work.
Several organizations are developing components for this approach.

Benchling
The company is connecting AI agents to experimental data, research context and scientific workflows, helping ensure that AI-generated recommendations can draw on relevant laboratory information.

Biolevate
Biolevate is integrating biological data, specialized scientific tools and agentic reasoning across drug discovery and development.

Boltz
Boltz is developing specialized models and tools for molecular structure prediction and candidate design. It is also previewing an agent intended to plan workflows, select models, interpret results and help guide the next experiment.

Automata
Automata is linking scientific agents with automated laboratory infrastructure so that experiments can be initiated and their results captured in structured, contextualized records.

Danaher and Genedata
Genedata’s AI co-scientist is designed to coordinate protein design and testing in a protein design foundry within Danaher’s autonomous laboratory. The workflow will incorporate RFD4-Proteina, using experimental findings to help shape subsequent design rounds.

Hologen
Hologen is combining its large medicine models with NVIDIA Nemotron to develop AI clinical scientists that analyze imaging and clinical data for insights relevant to patient outcomes and clinical trial design.

Therna Biosciences
Therna is using large language model reasoning to connect experimental evidence with RNA medicine design, aiming to help researchers translate biological findings into potential therapeutic candidates.
Together, these projects illustrate how AI could become part of the scientific workflow rather than remaining a standalone prediction tool. In an integrated system, computational design generates candidates, laboratory automation tests selected designs, and experimental results help determine what researchers should investigate next.
That feedback loop is a central ambition of autonomous scientific research. It could help teams investigate more possibilities, but the quality of the results will still depend on the underlying data, the reliability of the models and the rigor of experimental validation.
NVIDIA’s Healthcare AI Push Extends Into Hospital Operations
The announcement also covers the operational side of healthcare, where administrative workloads, fragmented information and limited specialist capacity can affect how efficiently patients receive care.
NVIDIA says Avelios, Corti, Dedalus, Dawn Health and Sword are building on its Nemotron models for applications ranging from AI medical scribes and AI-assisted clinical care to personalized patient support services.
PolyAI is taking a conversational AI approach through a multilingual deployment with medical software provider Corilus. According to NVIDIA, Corilus has a network spanning more than 70,000 healthcare professionals across Belgium and France. The collaboration is intended to help patients navigate pathways to care through conversational systems.
Other projects focus on the physical environment of hospitals. Apian and VitVio are using NVIDIA Cosmos world models and Omniverse libraries to develop digital twins — virtual representations of real-world systems — that can help monitor and improve hospital operations.
These applications broaden the focus beyond drug discovery and treatment planning. If developed and deployed effectively, AI systems could help healthcare organizations coordinate workflows, improve access to information and make better use of existing resources. Their practical value will depend on how reliably they perform in real clinical environments and how well they integrate with established care processes.
From AI Predictions to Testable Scientific Results
NVIDIA’s latest healthcare collaborations reflect a wider effort to connect generative AI with physical systems and specialized professional workflows.
RFD4-Proteina addresses the challenge of generating and evaluating new protein designs. The BioNeMo ecosystem is intended to connect those models with scientific agents and laboratory automation, while the Siemens Healthineers collaboration explores how AI and real-time computing could support more adaptive cancer treatment.
The projects are at different stages of development, and the announcement does not establish that they have already delivered new approved medicines, improved cancer survival rates or universally available autonomous healthcare services.
The immediate development is the release of RFD4-Proteina and the expansion of NVIDIA’s partnerships across research and healthcare. The longer-term test will be whether these systems can consistently turn computational predictions into experimentally verified discoveries and clinically useful improvements.
