AI Opening: NVIDIA BioNeMo Improves AWS Drug Discovery

Using Amazon Web Services, researchers and engineers at top pharmaceutical and techbio firms may now quickly and simply use NVIDIA Clara software and services for faster healthcare.

The program, which was unveiled at AWS re:Invent today, allows developers working in the healthcare and life sciences industry to integrate NVIDIA-accelerated products like NVIDIA BioNeMo, a generative AI platform for drug discovery that will soon be available on NVIDIA DGX Cloud on AWS. It is also currently accessible through the AWS ParallelCluster cluster management tool for high performance computing and the Amazon SageMaker machine learning service.

AWS is used by thousands of healthcare and life sciences businesses worldwide. With proprietary data, they will now have access to BioNeMo, enabling them to construct or modify foundation models for digital biology. Model training and deployment will be sped by the use of NVIDIA GPU-accelerated cloud servers on AWS.

Techbio innovators employing BioNeMo for generative AI-accelerated drug discovery and development include LabGenius, Alchemab Therapeutics, Basecamp Research, Character Biosciences, Evozyne, Etcembly, and AWS customers. They now have additional options to quickly scale up cloud computing resources for creating generative AI models that have been trained on biomolecular data thanks to this collaboration.

With this release, NVIDIA expands its portfolio of healthcare-oriented products on AWS, which includes NVIDIA Parabricks for accelerated genomics and NVIDIA MONAI for medical imaging processes.

NVIDIA BioNeMo: Introducing AWS and Advancing Generative AI for Drug Discovery

BioNeMo is a domain-specific framework for generative AI in digital biology that includes data loaders, pretrained large language models (LLMs), and optimized training recipes that can accelerate target identification, protein structure prediction, and drug candidate screening in computer-aided drug discovery.

Teams working on drug development can utilize BioNeMo to build or optimize models using their proprietary data, which can then be performed on cloud-based high performance computing clusters.

Using 256 NVIDIA H100 Tensor Core GPUs, one of these models the potent LLM ESM-2 achieves nearly linear scalability for protein structure prediction. Instead of taking a month to complete training, as stated in the original report, researchers may scale to 512 H100 GPUs and finish in a few days.

ESM-2 may be trained at scale by developers with checkpoints of 3 billion or 650 million parameters. The BioNeMo training framework supports other AI models, such as the protein sequence generation model ProtT5 and the small-molecule generative model MegaMolBART.

Using self-managed services like AWS ParallelCluster and Amazon ECS as well as integrated, managed services like NVIDIA DGX Cloud and Amazon SageMaker, BioNeMo’s pretrained models and optimized training recipes can help R&D teams build foundation models that can explore more drug candidates, optimize wet lab experimentation, and find promising clinical candidates more quickly.

NVIDIA Clara for Medical Imaging and Genomics is also accessible on AWS

With over 1.8 million downloads, Project MONAI, which NVIDIA cofounded and is enterprise-supported to support medical imaging workflows, may be deployed on AWS. Using their own healthcare datasets that are currently saved on AWS cloud services, developers may quickly annotate and construct AI models for medical imaging.

These models can be used for interactive annotation and fine-tuning for medical imaging segmentation, classification, registration, and detection tasks. They were trained on NVIDIA GPU-powered Amazon EC2 instances. Additionally, MRI image synthesis models included in MONAI can be used by developers to enhance training datasets.

In order to speed up genomics workflows, variant calling on the entire human genome can be accomplished with Parabricks in about 15 minutes as opposed to a day on a CPU-only system. Developers can easily scale up to handle massive volumes of genomic data over numerous GPU nodes on AWS.

AWS HealthOmics offers over twelve Parabricks workflows as Ready2Run workflows, allowing users to quickly run pre-configured pipelines.

Start accelerating AI workflows for drug development, genomics, and medical imaging with NVIDIA Clara on AWS.