US scientists have built a high-throughput biological platform that could speed up the search for rare-earth separation materials, a key step in strengthening supply chains for electronics, batteries, and magnets.

Researchers at Lawrence Livermore National Laboratory (LLNL) are using naturally occurring bacterial proteins to isolate and study rare-earth elements.

These proteins, known as lanmodulin, evolved in microbes that use rare-earth elements in their metabolism.

The goal is to turn this biological capability into a scalable tool for industrial metal separation.

But traditional protein screening methods are slow. Scientists typically test candidates one by one, making large-scale discovery impractical.

A new platform developed at LLNL changes that process by enabling parallel screening of hundreds of protein variants in a single run.

Spicy lambs system

The method, published in Nature Chemical Biology, is called SpyTag-Catcher Immobilization of Lanmodulin for Assaying Metal-Binding Selectivity, or SpyCI-LAMBS, nicknamed “spicy lambs.”

“It only took about a month to collect 600 proteins’ worth of data with this new assay,” said LLNL scientist and first author Patrick Diep. “It would have taken three to five years with the usual process.”

The system works by attaching engineered tags to proteins so they can automatically bind to a solid surface, eliminating the need for complex purification steps that usually slow down experiments.

Previous methods required researchers to extract proteins from bacterial mixtures containing thousands of other molecules, a process that significantly limited throughput.

“We started by just saying, ‘one by one, let’s go through these lanmodulin proteins and test them.’ We made it through a handful of them and realized it would take us the rest of our lives to effectively characterize them all,” said LLNL scientist and senior author Dan Park.

The new platform allows researchers to bypass that bottleneck entirely.

Proteins at scale

Using 96-well plates, scientists can now test dozens of lanmodulin variants in parallel against rare-earth elements. Multiple plates can also be run simultaneously, enabling large-scale mapping of metal-binding behavior.

The team identified eight distinct protein clusters with different rare-earth selectivity patterns.

One group of more than 200 variants showed improved performance in separating light rare-earth elements, a key challenge in refining critical materials.

Some engineered proteins were even able to complete separations in a single step, reducing process complexity.

The data generated is also being used to train machine learning models that predict how proteins will behave before they are physically tested.

Machine learning shift

By combining biology with data science, researchers aim to design proteins with targeted metal-binding properties rather than discovering them through trial and error.

“By transforming metalloprotein characterization from a low-throughput bottleneck into a scalable data-generation platform, the approach opens the door to predictive, data-driven design of metal-selective proteins,” said LLNL scientist and co-author Yongqin Jiao.

The platform may also extend beyond rare-earth elements, with researchers exploring applications in other critical metals.

“What I love about this approach is it reduces the cost of failure, so you can try some pretty wild ideas,” said Park. “We often throw in all kinds of different designs and concepts. And sometimes they work, sometimes they don’t, but the point is that we can afford to test these ideas now.”

The findings demonstrate how automation, biology, and machine learning are converging to reshape materials discovery at scale.

The study was published in Nature Chemical Biology.