Building the Agent-Driven Era of Science with NVIDIA BioNeMo Agent Toolkit

The Future of Science is Agent Driven.

To date, most AI-for-science efforts have confined themselves to narrow domains: protein folding, molecular generation, language models of nucleic acids. Lila's thesis is different. We are training a single reasoning model on data spanning across science. We post-train reasoning models, including the open source NVIDIA Nemotron 3, on scientific data that we generate in our AI Science Factories across life sciences, chemistry, and materials science. Reasoning models break complex problems into composable parts, and composability confers generality. A model that reasons across the broader areas of science can attack problems that would be impossible to solve in isolation.

We believe the next era of discovery will be driven by AI agents that reason, plan, and execute across entire scientific workflows. Rather than automating fixed workflows, our reasoning model iterates on experiments across scientific domains, turning the physical lab into a programmable extension of the model's reasoning. The Lila platform operates through scientific self-play with real-world experimentation: AI proposes the molecular or material hypotheses, then AI Science Factories – autonomous labs that run experiments –verify these hypotheses at scale.

BioNeMo augments scientific agents to understand real world data.

Biological discovery requires agents that understand molecular sequence, structure, and function. Lila trains the model to leverage scientific computational tools, including open-source BioNeMo models, proprietary Lila models, code-execution environments, and complex scientific simulations powered by NVIDIA ALCHEMI. The NVIDIA BioNeMo AgentToolkit provides domain-specific models, NVIDIA NIM microservices, open models, and life-science recipes built for biology. It augments our agent with the foundation to perform structure prediction and de novo design.

Lila is integrating BioNeMo to empower agents to make real world discoveries.

Lila's platform enables agents to develop new molecules and materials. A scientist can direct an agent to do in minutes or hours what could take a team weeks or months.

Agents can now run every step of the scientific method. Our agents use literature review tools powered by NVIDIA Nemotron Omni, draw on models we train on our own AI Science Factory data, and are integrated with NVIDIA BioNeMo models to make accelerated predictions for molecular performance. Most importantly, we enable AI to test hypotheses in the real world. These are often encoded in DNA and tested in our AI Science Factories to unlock the foundations of next-generation therapies.

Agentic science accelerates impact.

Lila's platform is optimized for the objective that ultimately matters: time from hypothesis to validated, functional molecule or material, while searching for breakthrough discoveries. A scientist who would have spent weeks designing and screening candidates can now direct an agent to propose, test, and refine hypotheses in a continuous closed loop, where the model designs the next batch of experiments, autonomous analysis updates the model with the latest results, and the cycle repeats without manual handoffs.  

This compresses the path from idea to computation to experiment, and it lets science at the pace of agents. The results already validate the approach. Lila's platform produced ultra sequences with direct application to next-generation CAR-T therapies, discovered novel catalysts for critical reactions, and designed novel protein binders. These are the kinds of discoveries that emerge when agents, grounded in real-world experimental feedback, are given the tools to act on biology and materials science rather than only describe it.

[About the author: Ben Kompa is a Co-founder and the Head of AI Lab Innovation at Lila Sciences.]

‍