NVIDIA Launches Large Language Models and Generative AI Services to Accelerate Life Sciences R&D

On March 21, 2023 (Pacific Time), NVIDIA launched a comprehensive suite of generative AI cloud services for customizing foundational AI models at GTC. These services will accelerate the creation of novel proteins and therapeutics, as well as advance research in genomics, chemistry, biology, and molecular dynamics.

As part of NVIDIA AI Foundations, the new BioNeMo cloud service for AI model training and inference can accelerate the most time-consuming and costly stages of drug discovery. Researchers can rely on it to fine-tune generative AI applications on their proprietary data, run AI model inference directly through a web browser or the new cloud application programming interface (API), and easily integrate into existing applications.

Kimberly Powell, Vice President of Healthcare at NVIDIA, stated, “The revolutionary power of generative AI has opened up vast prospects for the life sciences and pharmaceutical industries. NVIDIA’s long-term collaborations with pioneer enterprises in this field have driven the development of the BioNeMo cloud service. This service now functions as an AI-powered drug discovery laboratory, offering pre-trained models and enabling customization with proprietary data to support every stage of the drug development pipeline. This helps researchers identify appropriate therapeutic targets, design molecules and proteins, and predict their interactions within the human body, thereby facilitating the development of optimal drug candidates.”

Amgen, a world-leading biotechnology company, is already using this service to advance its research and development efforts.

Peter Grandsard, Executive Director of the Amgen Digital Innovation Research Acceleration Center and the Biologics Research & Development Department, stated: “BioNeMo has significantly accelerated our biologics R&D process. Leveraging this service, we can pre-train large language models for molecular biology on Amgen’s proprietary data, enabling us to explore and develop therapeutic proteins for next-generation medicines to help treat patients.”

BioNeMo cloud services include pre-trained AI models to help researchers establish AI-driven drug discovery pipelines. The service has been used by pharmaceutical R&D companies, including Evozyne and Insilico Medicine, to support the design of data-driven new candidate therapeutics.

Generative AI models can rapidly identify potential drug molecules and, in some cases, design compounds or protein-based therapeutics from scratch. Trained on large datasets of small molecules, proteins, DNA, and RNA sequences, these models can predict the 3D structures of proteins and the extent to which molecules dock with target proteins.

In addition to the previously announced MegaMolBART generative chemistry model, ESM1nv protein language model, and OpenFold protein structure prediction model, BioNeMo has now added six new optimized open-source models, including:

AlphaFold2：A deep learning model developed by DeepMind that can reduce the time required to determine protein structures from years to minutes or even seconds, using only the amino acid sequence of the proteins. This model has been used by over one million researchers.

DiffDock：To help researchers understand how drug molecules bind to target proteins, the model predicts the 3D orientation and anchoring interactions of small molecules with high precision and computational efficiency.

ESMFold：This protein structure prediction model leverages Meta AI’s ESM2 protein language model to predict the 3D structure of proteins from a single amino acid sequence, without requiring homologous sequence samples.

ESM2：This protein language model is used to infer machine representations of proteins, proving useful for downstream tasks such as protein structure prediction, property prediction, and molecular docking.

MoFlow：Used for molecular optimization and small-molecule generation, this generative chemistry model reconstructs molecules and proposes diverse chemical structures for potential therapeutic agents.

ProtGPT-2：This language model generates novel protein sequences, assisting researchers in designing proteins with unique structures, properties, and functions.

BioNeMo services enable users to easily access these generative AI models through a browser-based interface, facilitating interactive inference and visualization of protein structures. By combining BioNeMo with supercomputing resources on NVIDIA DGX Cloud, researchers can customize their models using fully managed software services powered by the NVIDIA Base Command platform and the NVIDIA AI Enterprise software suite.

Pharmaceutical companies and drug discovery startups are using BioNeMo and have achieved significant results multiple times.

Amgen used its proprietary antibody patent data to pretrain and fine-tune BioNeMo’s ESM model architecture. The company reduced the time required to train five custom models for molecular screening and optimization on DGX Cloud from three months to a few weeks.

Researchers at Evozyne, a Chicago-based biotechnology company and member of the NVIDIA Inception program, jointly developed with NVIDIA a deep learning model based on BioNeMo—the Protein Transformer Variational Autoencoder. This generative AI model was fine-tuned on Evozyne’s proprietary protein data, enabling researchers to design synthetic variants with performance significantly superior to that of naturally occurring enzymes.

Insilico Medicine, a senior member of the NVIDIA Inception program, is using BioNeMo to accelerate early-stage drug discovery. This process previously took more than four years and cost approximately $500 million. By leveraging end-to-end generative AI, Insilico identified a preclinical candidate drug in one-third of the time and at one-tenth of the cost. The drug is expected to enter Phase II clinical trials in patients soon.

The infiltration of generative AI into the healthcare sector has become an unstoppable trend. Following the initial introduction of generative AI for clinical documentation, NVIDIA has expanded its application scenarios to include drug discovery, a move expected to significantly enhance R&D efficiency. What other healthcare scenarios will adopt generative AI in the future? Only time will tell.

Apply for Early Access to BioNeMo.

Learn More at GTCAI and the Healthcare Industrylatest developments, including three sessions hosted by Amgen speakers, one session hosted by an Evozyne speaker, and another session on AlphaFold hosted by a DeepMind speaker. This year’s GTC is being held online and will continue through Thursday, March 23.Register here for free.

Watch the segment on BioNeMo Cloud Service in the GTC keynote address by Jensen Huang, Founder and CEO of NVIDIA.