Nvidia boosts generative AI for biology with BioNeMo

In September 2022, Nvidia announced the launch of its BioNeMo Large Language Model (LLM) service to help researchers build new artificial intelligence (AI) models for biology and it’s an effort that has yielded some strong early results.

Today, Nvidia, along with biotech startup Evozyne, are announcing that BioNeMo was used to help build a new generative AI model that could have a significant impact on helping improve human health as well as climate change. The generative AI model was used to create a pair of new proteins that are being detailed today. One of the proteins could one day be used to reduce carbon dioxide, while the other might help to cure congenital diseases.

>>Follow VentureBeat’s ongoing generative AI coverage<<

“Proteins are the building blocks of life,” Kimberly Powell, VP of healthcare at Nvidia, said during a press briefing. “The field of protein engineering looks to discover new proteins that can help us design more effective drugs, remove carbon from the air or make more environmentally friendly clothes.”

Generative AI is leading a revolution for healthcare

While generative AI has become a popular topic for discussion in recent months thanks to the popularity of services like ChatGPG, its application in the life sciences field is nothing short of revolutionary.

Earlier this week, life science vendor Absci announced that it has built a generative AI model using E. coli bacteria that can create entirely new antibody drugs. IBM has developed an open-source library called Generative Toolkit for Scientific Discovery (GT4SD) to help advance efforts as well.

Using generative AI to accelerate drug discovery is likely to be a major trend in 2023 overall as researchers look to reduce the time it takes to create new treatments.

Using Nvidia BioNeMo for generative AI

The Nvidia BioNeMo LLM is a model designed to help enable drug discovery. 

With generative AI, which uses large language models as a base, new things can be created based on data that the model has already been trained on.

Powell explained that the number of potential proteins far exceeds the number of particles in the universe. So in order to accelerate the discovery process, Evozyne and Nvidia used Nvidia BioNeMo to create the Prot-VAE model for protein discovery in just a few weeks.

She said that Evozyne used BioNeMo to fine-tune Prot-VAE for a specific protein family. Once it was fine-tuned, the new model has been used to generate an entirely new library of proteins that maintain or enhance a function in a specific family class of proteins.

“This model was able to invent proteins never before seen in nature,” Powell said. “This is the capability that is going to be how we can explore the infinite universe of proteins to discover new therapies in materials, energy sources and sustainable foods.”