Nvidia Delivers Generative AI for Healthcare Inference

At the recent Nvidia developer conference, GTC 2024, the tech giant introduced a groundbreaking development in the field of healthcare AI. Optimised packages of AI models and workflows with API have been bundled as NIMs (Nvidia Inference Microservices), providing developers with essential building blocks to create generative AI solutions for various healthcare applications, including drug discovery, medical technology, and digital health products.

With the unveiling of 25 NIMs, Nvidia is offering a diverse range of capabilities such as advanced imaging, natural language processing, speech recognition, digital biology generation, prediction, and simulation. These NIMs are designed to expedite the screening of drug compounds for discovery purposes and facilitate the collection of patient data in healthcare settings to aid in disease detection.

Kimberley Powell, Nvidia’s vice president of healthcare, emphasized the significance of NIMs in enhancing healthcare interactions between physicians and patients. She stated, “NIMs represent the fastest way to deploy AI-driven solutions in healthcare, empowering companies to fully leverage the potential of generative AI technologies to improve patient outcomes."

The NIMs are accessible through Nvidia AI Enterprise 5.0 software, offering specialized models tailored for drug discovery tasks. Notable examples include MoIMIM for generative chemistry, ESMFold for protein structure prediction, and DiffDock for analyzing interactions between drug molecules and their targets.

Running on the DGX cloud infrastructure, these NIMs deliver exceptional performance. For instance, the Vista 3D NIM enables the creation of precise 3D segmentation models, while the Universal DeepVariant NIM significantly accelerates genomic analysis workflows, outperforming CPU-based implementations by over 50 times.

In addition to the NIMs, Nvidia provides accelerated software development kits and tools like Parabricks, NeMo, and Metropolis, available as Nvidia CUDA-X microservices. These resources are invaluable for genomics analysis and medical imaging applications, further expanding the capabilities of developers in the healthcare AI domain.

One notable integration involves Cadence incorporating Nvidia’s BioNeMo microservices into its Orion molecular design platform to streamline drug discovery processes. Researchers utilizing Orion can efficiently generate, search, and model vast data libraries containing billions of compounds, enabling the creation of custom modules tailored to specific research requirements.