Genesis Workbench: A blueprint for industry AI in life sciences, powered by Databricks and NVIDIA

Summary

Genesis Workbench is an open blueprint for a Databricks-native life-sciences application that brings computational drug-discovery stages into one governed workbench. It combines Unity Catalog governance, MLflow tracking, Model Serving, serverless GPU compute, Databricks AI Search, and NVIDIA technologies including CUDA-X libraries, Parabricks, BioNeMo tools, GenMol, and Proteina-Complexa. Independent modules cover genomics, single-cell analysis, large- and small-molecule workflows, and model fine-tuning, with handoffs spanning gene-to-sequence resolution, structure prediction, docking, ADMET, and candidate ranking. A point-and-click React interface supports bench scientists, while declarative workflow generation and MCP exposure let pipelines and external clients use the workbench; inference runs on GPU endpoints inside the governed workspace without runtime external API dependencies. The stated aim is to let teams move from disease hypotheses to ranked therapeutic candidates on their own data, with a roadmap for automated workflow generation, BioNeMo Skills integration, and additional MCP services.

Context

Life-sciences R&D spans genomics, transcriptomics, structural biology, and chemistry, but these disciplines rarely share a common toolchain. The work involves highly sensitive and regulated assets, including sequences, compound libraries, assay results, and patient data, while drug discovery can take years and billions in investment.

Approach / What changed

Genesis Workbench uses a modular Databricks application with independently deployable scientific modules, Unity Catalog governance, MLflow tracking, GPU Model Serving endpoints, serverless GPU compute, and Databricks AI Search. A point-and-click React UI, declarative workflow generation, in-app handoffs, fine-tuning, and an MCP server connect bench scientists, computational workflows, models, and governed data.

Takeaways

• The workbench’s modules cover genomics, single-cell analysis, large-molecule design, small-molecule design, and NVIDIA BioNeMo model fine-tuning, with workflows connecting steps such as target structure prediction, docking, ADMET, and ranking.
• Models and data are downloaded into Unity Catalog, while inference runs on Model Serving endpoints in the organization’s workspace, removing runtime external-API dependencies described in the post.
• NVIDIA acceleration supports GPU variant calling with Parabricks, large-scale single-cell analysis with RAPIDS-singlecell, molecule generation with GenMol, and protein binder design with Proteina-Complexa, ProteinMPNN, and ESMFold.

Related reading