Artificial intelligence (AI) is changing how science gets done. At the U.S. Department of Energy's (DOE) Argonne National Laboratory, researchers are not just using AI as a tool. They are building AI systems that think, reason and experiment alongside scientists. Three new projects, funded in concert with DOE's Genesis Mission, put Argonne at the center of that transformation.
The Genesis Mission is a DOE initiative to harness AI to accelerate scientific discovery across disciplines. Argonne is uniquely positioned to lead this work. The laboratory combines world-class computing infrastructure, deep expertise in AI and machine learning, and hands-on biological research capabilities - including self-driving laboratories where robots and algorithms run experiments with minimal human intervention. That rare combination of strengths makes Argonne a natural home for projects that sit at the crossroads of AI and life science.
Each of the three projects - OPAL, IdeA and MELT-REE - tackle biological challenges in a complementary manner. Together, they represent a new model for how science can be done: faster, smarter and at a scale no human team could match alone.
OPAL: A Self-Driving Network for Biological Discovery
The goal of OPAL - Orchestrated Platform for Autonomous Laboratories to Accelerate AI-Driven BioDesign - is to build the infrastructure for self-driving laboratories across four national labs to run coordinated biological experiments autonomously.
Scientists Dion Antonopoulos, Ian Foster and Arvind Ramanathan helm Argonne's research and partner with DOE's Lawrence Berkeley National Laboratory (Berkeley Lab), Oak Ridge National Laboratory (ORNL) and Pacific Northwest National Laboratory (PNNL). Each laboratory contributes a distinct focus and set of capabilities. Argonne focuses on protein design - engineering the molecular building blocks of cells. Berkeley Lab and PNNL focus on studying microbial behavior. ORNL examines plant biology.
The vision is a system where an AI planning agent designs an experiment, dispatches tasks to the laboratory, monitors results and adjusts the next round of experiments.
One of Argonne's contributions to the project is humanoid robotics. Most laboratory automation relies on liquid-handling machines - devices that move and mix fluids with precision. But many experiments require more: adjusting instrument settings, responding to visual cues, operating equipment designed for human hands. Argonne is training humanoid robots - machines built to mimic human movement - to handle those tasks.
There are some experiments for which we need robots that can do specific types of experiments. We're pushing on that. We have humanoid robots in the lab where we are experimenting with what they can do and how we train them."
Arvind Ramanathan, Scientist, Argonne National Laboratory
OPAL also tackles the challenge of running experiments across institutions simultaneously. A single experiment might begin at Argonne, continue at Berkeley Lab using different instruments and scale across both sites at once. Building the software and coordination systems to make that work - reliably, at scale - is itself a major scientific achievement.
Argonne's efforts with its Autonomous Discovery initiative, its long-standing partnerships with the other national laboratories and its leadership in large-scale AI infrastructure make it a natural anchor for this multilaboratory effort. OPAL is not just a science project. It is a blueprint for how national laboratories can work together in the age of AI.
IDeA: Teaching AI to Think Like a Scientist
Designing a new enzyme - the biological molecule that catalyzes specific chemical reactions in living cells - can take a team of trained scientists a year or more to fully characterize it. The Intelligent Design Assistant for Enzyme Discovery and Biosynthetic Pathway Optimization, or IDeA, aims to compress that timeline to a matter of weeks.
Led by Ramanathan, IDeA is building a system of AI agents - think of them as specialized digital researchers - that can search millions of scientific papers, scan biological databases, compare molecular structures and generate hypotheses, all simultaneously. Where a human scientist might spend months combing through literature and databases to find a promising starting point, IDeA's agents can process roughly 3 million scientific documents in about a week on a supercomputer.
The first target is enzymes that produce nylon-like biopolymers - materials used in manufacturing. Finding and optimizing the right enzyme today requires painstaking literature review, database searches and laboratory testing, often spread across an entire research team. IDeA automates and parallelizes that workflow, allowing many lines of inquiry to run at once and converge on answers far more quickly than any single team could manage.
The project also addresses a deeper challenge: building AI that can reason about biology, not just retrieve information. Argonne is training new biological reasoning models using data generated in its own labs. When AI agents disagree with each other - a real risk when multiple systems run independently - the system is designed to resolve conflicts and stay grounded in established science, ensuring results are trustworthy and reproducible.
"Argonne hosts some of the best bioinformatics data sets on the planet," Ramanathan said.
Argonne's data assets, its deep experience building large-scale AI systems and foundation models for biology, and the raw computing power to run them give the laboratory capabilities that are difficult to match anywhere else. IDeA's experimental partner is ORNL, which generates the biological data the AI system learns from - closing the loop between computation and the lab bench.
MELT-REE: Using Microbes to Recover Rare-Earth Elements
Rare-earth elements are essential to modern technology. They're found in virtually every electronic device, including advanced defense systems. The U.S. has significant deposits of these materials, but extracting them is costly and chemically intensive. MELT-REE - Multimodal Engineering and Leaching Technology for Rare-Earth Extraction - is exploring a biological alternative.
The project uses microbes, specifically bacteria, to leach rare-earth elements from solid waste materials, such as mine tailings - the leftover rock and debris from mining operations - and electronic waste. This process, called bioleaching, already works at small scales. The challenge is making it fast, robust and efficient enough for industrial use.
"Conventional chemical leaching works, but it comes at a cost - corrosive reagents, significant energy input and acid waste that has to go somewhere," said Argonne bioscientist Daniel Schabacker, who co-leads the project with Ramanathan. "Biology offers a fundamentally different approach: a microbe that does the same job without the chemical footprint."
Several barriers stand in the way. The bacteria generate acid as part of their leaching activity. Yet that same acid, once it accumulates, begins to inhibit the bacteria themselves. They also struggle when metal concentrations in their environment rise, and they become inefficient when the amount of solid feedstock exceeds about 1% by weight. Industrial processes need that figure to reach 10% or higher.
MELT-REE attacks these barriers on two fronts. In the lab, Schabacker's team is screening a collection of 2,733 bacterial strains built by collaborators at Cornell University - each with a different gene disabled - to identify which genes control leaching ability. That work uses Argonne's self-driving laboratory, an automated system that runs experiments far faster than traditional methods.
The data flows directly to Ramanathan's computational team, which uses AI to identify patterns, predict which biological pathways matter most and suggest ways to engineer better-performing microbes. The goal is a microbe optimized for industrial-scale rare-earth extraction and recovery - tougher, faster and more productive than anything found in nature.
Argonne's combination of self-driving labs, AI expertise and high performance computing makes it one of the few places in the world where this kind of integrated, AI-guided biological engineering is possible.
The OPAL, IdeA and MELT-REE projects are funded by the DOE's Office of Science, Biological and Environmental Research program.
Deep-Learning Holography Detects Elusive Circulating Tumor Cells in Prostate Cancer