Deploy RFD4-Proteina on Databricks

RFD4-Proteina, co-developed by the University of Washington Institute for Protein Design and NVIDIA, is now available on Databricks through Genesis Workbench on AI Runtime. The model, part of NVIDIA’s BioNeMo open model suite, enables researchers to transform protein design ideas into testable designs using AI-assisted development and serverless NVIDIA GPU compute. The integration provides a unified environment to connect protein-design workflows with existing research data, reducing the need for separate infrastructure setups.

The integration simplifies model deployment by embedding RFD4-Proteina into Genesis Workbench, which offers prebuilt workflows for tasks such as antibody and vaccine design. Researchers can access the model through modular blueprints that illustrate how to prepare inputs, evaluate candidates, and analyze results within Databricks. Teams can store and govern model weights via Unity Catalog, ensuring reproducibility and collaboration across experiments.

Genesis Workbench supports natural language-driven development through Databricks Genie Code, which helps researchers build, debug, and adapt workflows. NVIDIA BioNeMo Agent Toolkit skills provide specialized instructions for biological models, including structure prediction and protein design, while Claude-based coding agents offer additional guidance for integrating new models like RFD4-Proteina. The environment allows customization to align with specific research requirements and data conventions.

Databricks AI Runtime (AIR) provides serverless NVIDIA GPU compute, enabling teams to run workloads without provisioning or maintaining separate GPU clusters. The documented setup includes selecting an NVIDIA H100 accelerator and supported environment, reducing infrastructure overhead. Teams can adapt BioNeMo Recipes for training or fine-tuning on proprietary data, while MLflow experiment tracking and Unity Catalog governance ensure repeatable workflows and enterprise controls.