Nvidia forges trio of AI-based drug discovery partnerships with AstraZeneca, Schrödinger, University of Florida

NVIDIA, world’s 22nd fastest supercomputer, DGX SuperPOD, supercomputer
Three new collaborations involving Nvidia will rely on advanced artificial intelligence technology for largely automated computational drug discovery. (Nvidia)

They say two heads are better than one, but Nvidia is betting four heads might be the best option for supercharging computational drug discovery.

The chip maker has established three new partnerships with the same goal in mind: furthering advanced artificial intelligence research technology in the drug discovery process.

Two of the collaborations—with AstraZeneca and the University of Florida—will incorporate transformer-based neural network architectures. This relatively new technology, a form of natural language processing, learns a language’s grammar and syntax as it works rather than requiring extensive pre-training by researchers.

In the same way, the technology can master the “language” of biochemistry, allowing it to search massive data sets for potential molecular targets and predict how they might react to certain chemicals, all without being fed manually annotated examples before beginning processing.

RELATED: Pfizer the latest pharma to befriend Iktos for its AI drug discovery platform

With the Big Pharma, Nvidia is developing a model for reaction prediction, molecular optimization and de novo molecule creation called the MegaMolBART model. It will be built on the partners’ transformer frameworks—Nvidia’s Megatron and AstraZeneca’s MolBART—and trained on the public-access ZINC database of chemical compounds.

Through that training, the model will gain a thorough understanding of chemical structure, which it will use to predict chemical reactions and devise novel molecular structures that could potentially become new drug candidates.

Once complete, MegaMolBART will be offered as an open-source model, available to all researchers via Nvidia’s NGC software catalog. It will also be one of the first projects processed using Cambridge-1, the U.K.’s most powerful supercomputer, which is currently under development by Nvidia, AstraZeneca, GlaxoSmithKline and the National Health Service.

RELATED: Insitro raises $400M for machine learning-powered drug discovery efforts

Meanwhile, at the University of Florida, the new GatorTron model is being built on Nvidia’s BioMegatron transformer and trained on more than 2 million UF Health patients’ records, representing more than 50 million clinical interactions.

The resulting model will have both clinical and pharmacological uses. Through its ability to identify the best candidates for clinical trials and predict life-threatening conditions, GatorTron will simultaneously speed up the drug-testing process, from clinical trial recruitment to analysis of the trials’ results.

“GatorTron leveraged over a decade of electronic medical records to develop a state-of-the-art model,” Joseph Glover, the university’s provost, said. “A tool of this scale will enable healthcare researchers to unlock insights and reveal previously inaccessible trends from clinical notes.”

RELATED: UCB taps Microsoft to accelerate drug discovery, clinical trials

A third AI-based partnership announced this week will see Nvidia join forces with digital molecule designer Schrödinger to enhance the latter’s existing computational drug discovery platform. The pair will make Schrödinger’s software compatible with Nvidia’s supercomputer systems, resulting in even faster and more accurate molecular modeling.

The platform will be available to any pharmaceutical company with access to Nvidia’s newly unveiled DGX SuperPOD cloud-based supercomputer.

“Computational drug discovery is improving accuracy with Schrödinger’s advanced combination of simulation with machine learning,” Kimberly Powell, Nvidia’s vice president of healthcare, said. “Together, we’re giving the pharmaceutical industry a scientific instrument that delivers super-high-throughput lead generation to improve and accelerate the success of drug candidates.”