NVIDIA Explores AI-Assisted Adaptive Radiotherapy With Siemens Healthineers

NVIDIA is exploring AI-assisted adaptive radiotherapy with Siemens Healthineers, adding treatment-planning research to its health care AI work as it also introduces a separate protein-design model.

At GTC Berlin on Oct. 21, NVIDIA announced the exploratory collaboration with Siemens Healthineers to develop adaptive radiotherapy. The work centers on agent-assisted treatment planning, patient-specific simulations and real-time computing. NVIDIA Nemotron models are being considered for planning support. CUDA could support dose calculations, while Holoscan could connect live sensor data with AI and simulation to support treatment adjustments. Arthur Kaindl, head of Varian, Siemens Healthineers’ cancer-care business, said “treatment should adapt to the patient.”

Separately, NVIDIA announced RFD4-Proteina, a protein-design model developed with the University of Washington’s Institute for Protein Design. The model generates protein sequences and structures, which researchers can search and refine using molecular inputs, constraints and scoring objectives. It is available under a BSD-3 open-source license. David Baker, director of the institute, said, “RFD4-Proteina expands what scientists can create.”

Across the computational binding test, the model’s clustered success rate was 10.6%, versus 5.9% for the runner-up, and its throughput was 42 times higher. These results measure computational design performance. In separate lab tests, 12 of 18 designed calmodulin binders showed apparent binding below one micromolar. Those 12 designs showed no detectable binding without calcium.

NVIDIA also named Automata and Benchling among companies building scientific-agent workflows. Automata is connecting agents with laboratory equipment, while Benchling is supplying agents with experimental data and research-process context. RFD4-Proteina is part of NVIDIA’s broader open AI model lineup.