Nvidia unveils generative AI programs for drug molecule design, protein predictions

A little more than a year after announcing plans for its molecule design programs to set up shop in the field of generative AI, Nvidia is opening its doors. 

Just as call-and-response apps like ChatGPT are trained on the various connections between words of the English language, the company’s BioNeMo toolkits can now draw lines among the DNA, RNA and proteins that ultimately govern cell function, according to Nvidia—with the goal of helping drug discovery researchers to conduct more experiments virtually before pursuing them in the physical world.

Nvidia plans to offer the expanded BioNeMo models as a series of cloud-based enterprise services—with predictors for drug molecule binding successes and changes in protein structures—hosted through its AWS HealthOmics platform. The announcement was made during Nvidia’s GTC conference this week in San Jose, California.

The new AI programs include DNABERT, a foundation model trained on genomic sequences that aims to predict the function of different regions of genetic code and the effects of mutations. Another upcoming model is scBERT, trained on single-cell RNA sequencing data for exploring gene knockout hypotheses. Meanwhile, EquiDock looks to forecast how two proteins will interact with each other in 3D. 

Nvidia said that more than 100 biopharma and AI drug discovery companies already employ its BioNeMo platform, including Amgen, Astellas Pharma, Cadence, Iambic, Insilico Medicine, Recursion and Terray Therapeutics. 

"Generative AI is transforming drug discovery by allowing us to build sophisticated models and seamlessly integrate AI into the antibody design process,” said Amgen’s chief technology officer David Reese. “Our team is harnessing this technology to create the next generation of medicines that will bring the most value to patients.”