Nvidia Corp. continues to expand its Clara healthcare platform with the addition of computational drug discovery and medical imaging tools based on its DGX A100 platform, related InfiniBand networking and its AGX developer kit.

The Clara partnerships announced during this week’s Nvidia GPU Technology Conference include development kits, pre-trained training models that include an emerging neural network framework and a DGX-based “SuperPOD” supported by Mellanox HDR InfiniBand switching.

The tools hew to the overarching theme of GTC21: “We have to make AI easier to use,” Nvidia CEO Jensen Huang declared in his keynote address.

Those new tools support the GPU leader’s healthcare initiative dubbed Clara Discovery, a collection of AI frameworks, applications and pre-trained models aimed at accelerating drug discovery. Along with collaboration with pharmaceutical research specialist Schroedinger, Nvidia said Monday (April 12) it also will work with drug maker AstraZeneca and the University of Florida on drug discovery. It is also working on device applications for its embedded AI platform with Carestream Health, a medical imaging specialist, and others.

Huang noted the combination of AI and pharmaceutical research has resulted in generative models able to define new drugs in less than two years. That process currently takes up to a decade.

Generative models are used to read the structure of promising drug compounds to generate potentially effective, novel compounds. “Now we can use AI to generate candidate compounds that can then be further refined with physics-based simulations,” Huang said.

It takes about 8 seconds to generate a single molecule using a CPU; Nvidia claims its GPU platform can render a molecule via a single A100 in about 0.3 seconds. Those steps could then be scaled via the DGX SuperPOD to generate what Huang said would be thousands of molecules per second.

Collaboration with Schroedinger will combine the DGX A100 platform with the chemical simulation software specialist’s machine learning techniques for accelerating drug discovery. The evaluation of tens of thousands of molecules for each potential drug candidate currently requires hundreds of thousands of hours of GPU time on HPC platforms.

Schroedinger is already a heavy user of Nvidia GPUs, recently entering into an agreement to use hundreds of millions of Nvidia GPU-hours on the Google Cloud, Huang said.

For drug researchers unable to use the cloud, Nvidia said it will help accelerate Schroedinger’s drug discovery workflow via Clara Discovery libraries and DGX. Schroedinger’s platform is used by some of the largest drug makers. “Their researchers are going to see a giant increase in productivity,” Huang asserted.

Among the goals of the collaboration is accelerating computing-intensive drug discovery at “supercomputing scale,” the partners added. That would allow drug makers to simulate molecular combinations with AI and physics to identify promising compounds for further investigation.

Another Clara initiative will leverage transformer-based neural networks that allow drug investigators to sift through massive data sets using “self-supervised” training methods. That approach eliminates the need to manually label examples during pre-training.

Collaboration with AstraZeneca will help accelerate the drug maker’s generative AI model for drug discovery based on transformer networks. Those emerging networks are trained for inference on sequential data. The effort will be among the first to run on Cambridge-1, the U.K.’s largest supercomputer based on Nvidia’s DGX A100 processor.

“Our aim is that neural networks trained on molecular structure data will be able to learn the relationships between atoms in real-world molecules,” said Ola Engkvist, head of molecular AI at AstraZeneca. The model will be released as open source to the drug investigators, the partners added.

Meanwhile, the University of Florida will contribute its implementation of Nvidia’s Megatron framework called GatorTron along with a pre-trained model called BioMegatron. Both are available on Nvidia’s NGC Software Catalog.

Nvidia also announced a collaboration with medical imaging specialist Carestream Health to incorporate its Clara AGX embedded AI platform into imaging devices used for X-ray screening. Medical imaging is seen as an early application for machine vision algorithms, helping to eliminate issues such as confirmation bias, industry sources note.

Carestream will utilize the Clara AGX platform for scanning both single-frame and streaming X-rays, Nvidia said. An AGX developer kit now generally available includes AI processing and 3D visualization. The kit focuses on development of software-defined instruments ranging from microscopes and ultrasound to endoscopes.

Farther afield, Nvidia is also working with developers incorporating AI and computer vision into surgical applications. “Surgical intelligence” specialist Theator said it is working with Nvidia to process unstructured video data from surgical procedures to improve compute vision models used to assist surgeons.