Nvidia Introduces Agentic AI to Transform Biotech Discovery

Artificial intelligence was a focal point at the Bio International Convention in San Diego, the largest biotech event featuring a wide range of industry vendors. During a special address, Kimberly Powell, vice president and general manager of healthcare and life sciences at Nvidia Corp., argued that agentic AI is poised to transform biotech in a manner similar to its impact on software. Nvidia's BioNeMo is the framework that enables generic large language models to function as "AI scientists" that are both faster and more cost-effective.

Powell outlined the current state of the industry, stating, "We are witnessing the fastest platform shift the life sciences industry has ever seen." She likened AI to groundbreaking scientific instruments such as the microscope and gene sequencing, noting that this new technology not only observes but also reasons, plans, and acts.

At the convention, Nvidia introduced its BioNeMo Agent Toolkit, a software stack that converts large language models into domain-specific AI agents capable of executing comprehensive biology and chemistry workflows, from literature reviews to protein design and lab automation, while optimizing for performance and cost.

Powell asserted that the life sciences sector, which has a $300 billion annual pharmaceutical budget and a global R&D expenditure reaching $3.8 trillion, has been preparing for this shift for a decade. There has been a surge in AI research across various fields, and Nvidia has been developing the necessary infrastructure, including GPUs and specialized libraries.

The emergence of agentic AI in the past 12 to 18 months has been significant, with systems that integrate a large language model "brain" with tools, memory, security policies, and multistep workflows. Nvidia's BioNeMo Agent Toolkit is tailored for life sciences, enabling companies to build specialized agents that can coordinate tasks across digital and physical labs.

Powell highlighted the toolkit's ability to enhance cost-performance efficiency, allowing agents to complete complex workflows with improved accuracy and reduced computational expenses. The BioNeMo toolkit is designed to be agent-agnostic, supporting various AI models, which is crucial for organizations seeking flexibility in their drug discovery processes.

To illustrate the toolkit's capabilities, Powell described a protein-binder design workflow targeting MCL1, a protein associated with tumor cell survival. Traditionally, this process requires extensi