Nvidia packages inference to deliver generative AI for healthcare

Optimised packages of AI models and workflows with API have been packaged as NIMs (Nvidia Inference Microservices) which developers can use as building blocks to develop generative AI for healthcare, from drug discovery, med-tech and digital health products.

Nvidia announced 25 NIMs at its developer conference, GTC 2024, offering advanced imaging, natural language and speech recognition, digital biology generation, prediction and simulation.

They can be used to accelerate screening of drug compounds for drug discovery as well as in healthcare practices to gather patient data for disease detection.

Kimberley Powell, Nvidia’s vice president of healthcare explains that NIMs are the “fastest way to deploy healthcare interactions between the physician and the patient.

“By helping healthcare companies easily build and manage AI solutions, we’re enabling them to harness the full power and potential of generative AI,” she said.

The NIMs are available through Nvidia AI Enterprise 5.0 software. Models for drug discovery include MoIMIM for generative chemistry, ESMFold for protein structure prediction and DiffDock to understand how drug molecules interact with targets.

All NIMs run in the DGX cloud. Other NIMs are the Vista 3D for the creation of 3D segmentation models and the Universal DeepVariant for genomic analysis workflows which is over 50x faster than a DeepVariant implementation running on a CPU.

Accelerated software development kits and tools, (e.g., Parabricks, NeMo and Metropolis) can also be accessed as Nvidia CUDA-X microservices for use in genomics analysis or medical imaging.

Cadence is integrating Nvidia’s BioNeMo microservices into its Orion molecular design platform to accelerate drug discovery. Researchers using Orion can generate, search and model data libraries with hundreds of billions of compounds. This enables researchers to generate modules optimised for specific needs in a research programme.