NVIDIA Launches Open-Source BioNeMo Framework at SC24 to Scale Digital Biology Research Across Global Biopharma and Scientific Communities

Argonne National Laboratory, Flagship Pioneering, Terray, Weights & Biases, and dozens of other organizations are driving the advancement of biomolecular science.

NVIDIA today announced that leaders in the global pharmaceutical and tech-bio industries, academic pioneers, and AI researchers are using open-sourceNVIDIA® BioNeMo™framework to advance drug development and accelerate molecular design.

Researchers require specialized biomolecular models and datasets to gather insights at scale, enabling faster design of therapeutic regimens. The open-source BioNeMo framework provides a suite of accelerated computing tools designed to deliver exponentially scalable AI models for biomolecular research, bringing a new level of supercomputing power to the biopharmaceutical industry.

Kimberly Powell, Vice President and General Manager of Healthcare and Life Sciences at NVIDIA, stated, “The recent Nobel Prize in Chemistry demonstrates that the convergence of AI, accelerated computing, and expanding datasets has created unprecedented opportunities for the pharmaceutical industry. To help unravel the complex mysteries of biological systems, we have launched the BioNeMo open-source framework, which will enable researchers worldwide to develop life-saving therapies more rapidly.”

Leading biotechnology companies, computational biology researchers, and AI platform companies and organizations have contributed or are poised to contribute to this open-source framework, including A Alpha Bio, Argonne National Laboratory, Dyno Therapeutics, Genentech (a member of the Roche Group), Ginkgo Bioworks, Relation, VantAI, and Weights & Biases. Additionally, leading key research institutions renowned for advancing computational science are leveraging the BioNeMo framework to drive innovation.

Arvind Ramanathan, head of the Computational Science Division at Argonne National Laboratory, stated, “Argonne National Laboratory has contributed billion-parameter biological models, which require specialized software for training in high-performance computing environments. BioNeMo provides an enterprise-grade open-source solution for Argonne National Laboratory and the broader biotechnology community, enabling researchers to easily scale up the training of large biological foundation models even in laboratories that lack sufficient computational expertise.”

This end-to-endNVIDIA BioNeMoThe platform is specifically designed to accelerate the creation, customization, and deployment of AI-driven drug discovery and molecular design models. By seamlessly integrating with accelerated computing infrastructure, it reduces costs, enables scalability, and expedites the drug discovery process, thereby extracting rapid and reliable insights from biomolecular data.

In addition to the BioNeMo framework, the platform also supportsNVIDIA NIM™(Optimized secure, scalable AI inference microservices) andNVIDIA BioNeMo Blueprints, the latter is an optimized reference design for wet lab and computational workflows.

NVIDIA has also released a suite of optimized, easy-to-use new NIM microservices for BioNeMo. These microservices can be deployed quickly and easily on-premises, in any data center, or in the cloud, enabling developers to flexibly run applications across diverse environments and accelerating the time from inference to insight in biopharmaceutical R&D.

The newly launched NIM microservices support industry-leading models, including:

• AlphaFold2 It is a deep learning model developed by Google DeepMind, designed specifically to revolutionize protein structure prediction. By using the AlphaFold2 NIM microservice for BioNeMo, researchers have achieved a five-fold increase in the speed of predicting protein structures in near real-time.

  
  

• DiffDock 2.0 Built on research from the Massachusetts Institute of Technology and trained on the gold-standard PLINDER dataset, the DiffDock 2.0 NIM microservice enables researchers to predict molecular orientation 6.2 times faster with a 16% improvement in accuracy.

  
  

• RFdiffusionandProteinMPNN NIMMicroservices help researchers design novel proteins that bind to target molecules more rapidly, thereby accelerating the development of new protein-based therapeutics.

BioNeMo now incorporates new acceleration libraries, including cuEquivariance, to accelerate the mathematical computations essential for DiffDock chemical predictions.

Enterprises can leverage NVIDIA BioNeMo Blueprints, a customizable catalog of reference AI workflows, to help developers scale their AI deployments into enterprise-grade production pipelines.

The BioNeMo Blueprint for virtual screening provides a customizable and easy-to-follow workflow that not only leverages NIM microservices to accelerate small-molecule design but also saves time and costs.

More than 200 tech-biotech companies, large pharmaceutical firms, and startups have already integrated BioNeMo into their computer-aided drug discovery platforms and workflows.

Global systems integrators, software providers, and cloud service providers such as Accenture, Amazon Web Services (AWS), and Deloitte are bringing NVIDIA BioNeMo Blueprints to enterprises worldwide.