Argonne Launches Three DOE-Funded AI Biology Projects

Argonne National Laboratory is leading three new DOE-funded projects that put AI closer to the scientific bench: one would coordinate autonomous experiments across four national laboratories, another would search for useful enzymes, and a third would engineer bacteria to recover rare-earth elements from waste.

The projects are part of the Energy Department’s Genesis Mission and are funded through the DOE Office of Science’s Biological and Environmental Research program. They share an ambition to shorten the loop between a computer’s hypothesis and a laboratory result, but they tackle sharply different bottlenecks in biological research.

One network, three approaches

OPAL, short for Orchestrated Platform for Autonomous Laboratories to Accelerate AI-Driven BioDesign, is the infrastructure play. It aims to link Argonne with Lawrence Berkeley, Oak Ridge and Pacific Northwest national laboratories in a coordinated autonomous-lab network. An AI planning agent would design an experiment, send work to laboratories, monitor results and revise the next round.

That makes OPAL less a single discovery engine than an operating system for experiments spread among institutions. Argonne is also experimenting with humanoid robots for work involving equipment made for people, rather than only liquid-handling machines. The difficult part is not simply automating a step; it is making a distributed sequence of experiments respond reliably to changing results.

The search problem versus the scale problem

IDeA, the Intelligent Design Assistant for Enzyme Discovery and Biosynthetic Pathway Optimization, focuses on the information-heavy front end of biological discovery. Its AI agents are intended to examine papers, databases and molecular structures at the same time, then generate hypotheses for researchers to investigate. The initial target is enzymes that produce nylon-like biopolymers used in manufacturing.

Argonne says the system can process roughly 3 million scientific documents in about a week on a supercomputer. Speed alone is not its stated goal: Argonne is training biological reasoning models on laboratory-generated data and designing ways to resolve disagreements among agents while keeping results grounded in established science.

What each project is trying to change

• OPAL would coordinate experiment planning and execution across four laboratories.
• IDeA aims to compress the search for enzyme candidates by parallelizing research tasks.
• MELT-REE seeks microbes that can make rare-earth recovery more viable at industrial scale.

A biological constraint that software cannot erase

MELT-REE faces a more physical test. Bioleaching already works at small scales, but the bacteria can be inhibited by the acid they produce and by rising metal concentrations. Argonne says they also become inefficient when solid feedstock exceeds about 1% by weight, while industrial processes need 10% or more.

Researchers are screening 2,733 bacterial strains, each with a different gene disabled, to identify genes that govern leaching ability. The resulting data is meant to feed an AI system that identifies useful patterns and suggests which biological pathways could produce tougher, faster microbes. It is an effort to use AI not as a substitute for bench work, but as a way to choose the next experiment more effectively.