DomainHuntdomainhuntAdvertise

Los Alamos and NVIDIA Unveil Chelatron, a Super Intelligence System for Molecular Discovery

Los Alamos National Laboratory and NVIDIA have developed Chelatron, a Super Intelligence-powered system that designs and screens molecules for cancer treatment and plutonium purification. Its first campaigns examined nearly 65,000 possible molecules.

GovernmentLos Alamos Reporter ·

Los Alamos National Laboratory and NVIDIA have developed a Super Intelligence-powered system that can design and evaluate new molecules, potentially accelerating discoveries in cancer treatment and nuclear materials processing. Called Chelatron, the system uses computational models to generate chemical candidates, predict their performance and refine subsequent designs. Its initial research has already examined almost 65,000 possible molecules.

The project addresses a longstanding problem in chemistry: finding molecules capable of distinguishing between elements that behave almost identically. Los Alamos scientist Ping Yang, recently named a 2026 Laboratory Fellow, led the effort to develop a more efficient approach. Rather than relying exclusively on existing chemical knowledge and limited experimental datasets, her team worked with NVIDIA to build a system that generates new molecular possibilities and evaluates them computationally.

AI tackles molecular discovery

The central challenge involves chelators, specialized molecules that attach strongly to metal ions. These compounds can capture particular metals, allowing scientists to separate them from mixtures, transport them or prevent unwanted chemical reactions. Their usefulness depends on selecting the right molecular structure for the element being targeted.

That distinction becomes particularly important when working with the f-elements, a family encompassing lanthanides and actinides. Lanthanides include rare-earth elements used in electronics and magnets, while actinides include uranium and plutonium. Members of these groups often have remarkably similar chemical properties, making selective separation difficult.

Yang illustrates the problem through an unusual comparison. She shows colleagues a picture pairing a Chihuahua with a blueberry muffin. From a distance, the dog's facial features resemble the berries on the muffin. Her analogy compares europium with americium, two elements that can occur in spent nuclear fuel and radioactive waste. Their chemical similarities make separating them a demanding task.

Scientists sometimes need to isolate americium from other elements before nuclear waste can be stored. Conventional separation methods can require considerable time and effort because the target element must be distinguished from substances with closely related chemical behavior.

"We needed to be smarter."

— Ping Yang, Los Alamos scientist

Why existing chemistry struggles

Chelation has been used in science for more than a century. Its applications extend well beyond nuclear materials, including the purification of rare-earth metals, water treatment, food production, cosmetics manufacturing and medicine. In cancer care, chelators can help attach radioactive isotopes to molecules designed to deliver them to specific biological targets.

Despite this long history, designing improved chelators remains difficult. Researchers must consider different combinations of atoms, molecular arrangements and chemical properties. Each possible structure represents another candidate that may need to be investigated before scientists can determine whether it performs the intended task.

Yang says the traditional approach to separating f-elements had seen little fundamental progress since the 1970s. One reason is the enormous number of possible molecular designs. Researchers have historically had relatively little information about which structures might work because investigating them individually consumes substantial time and resources.

This creates a problem for conventional computational research, too. Many approaches depend on extensive existing datasets to train models and guide predictions. Yang's team faced a different situation: the available information about potential chelators was incomplete, and the researchers wanted to discover possibilities that had not already been explored.

How Chelatron works

The Los Alamos researchers approached NVIDIA with a proposal to combine its reasoning model with models similar to those used in pharmaceutical research. The objective was to create a system capable of proposing new molecular structures rather than simply searching through existing chemical information.

The resulting workflow connects several computational stages. First, the system generates ideas for possible chelator designs. Models then turn those proposals into candidate molecular structures. Computer simulations, effectively virtual chemistry experiments, estimate how the candidates might behave. Their results return to the models, helping refine the next round of proposals.

The process creates a feedback loop in which each computational investigation supplies information for subsequent decisions. Instead of treating molecular design as a sequence of isolated experiments, the system repeatedly proposes, evaluates and improves possible solutions.

Yang and her colleagues call this an agentic SI-based system. Agentic systems can coordinate multiple steps toward a defined objective rather than responding only to an individual instruction. In this case, the system carries out a research workflow resembling the successive decisions a human scientist makes when investigating a new chemical compound.

The researchers named the resulting molecular discovery capability Chelatron. For Los Alamos, the main advantage is speed: computational screening can investigate many more possible structures than conventional laboratory experimentation could practically examine one by one.

Cancer treatment provides first test

The team initially demonstrated Chelatron through research into targeted cancer treatment. Specifically, researchers sought improved chelators for actinium-225, a radioactive isotope used in a medical approach called targeted alpha therapy, or TAT.

Targeted alpha therapy aims to deliver radioactive material directly to cancer cells. The isotope releases alpha radiation capable of destroying targeted cells while limiting exposure to surrounding healthy tissue. Achieving that selectivity requires an effective delivery mechanism, making the chemical properties of the chelator particularly important.

A suitable molecule must hold the radioactive isotope securely and help deliver it to the intended biological location. If the binding performance is inadequate, the delivery process becomes more difficult to control. Developing better compounds could therefore support improvements in the precision of this treatment approach.

Chelatron's first two research campaigns produced 119 design proposals and investigated almost 65,000 molecular possibilities. The researchers subsequently generated and evaluated three-dimensional models for approximately 2,500 candidates, narrowing the investigation to compounds predicted to offer promising performance.

The computational assessment identified 679 molecules predicted to bind actinium at least as effectively as, or more effectively than, the leading chelator selected as a comparison. These results provided the researchers with a much smaller group of possibilities for further investigation.

119molecular design proposals
65,000possible molecules examined
2,500three-dimensional molecular models evaluated
679candidates predicted to match or outperform the comparison chelator

Laboratory testing begins

Computational predictions are not the same as demonstrated chemical performance. The next stage therefore moves the most promising candidates out of simulations and into physical experiments.

Stosh Kozimor, a Los Alamos scientist and Laboratory Fellow, is leading this validation phase. Yang's team has passed him the five highest-ranked chelators identified by Chelatron. His group will synthesize the proposed compounds and test their actual chemical behavior in the laboratory.

Producing specialized molecules can be technically demanding. Nevertheless, Kozimor sees an important benefit in using computational discovery to identify structures that might not emerge from conventional chemical intuition. The system can direct researchers toward possibilities they might otherwise overlook, giving experimental scientists new candidates to investigate.

The distinction matters because the value of the technology ultimately depends on whether its predictions survive laboratory testing. A molecule that appears promising in a simulation must still be synthesized and shown to perform as expected. Both Yang and Kozimor emphasize that experimental validation remains essential.

Yang says Chelatron can screen hundreds of new chelators overnight. That capability changes the balance between computational investigation and physical experimentation: researchers can use simulations to narrow a much larger field before committing laboratory resources to a handful of compounds.

Plutonium purification is another target

The cancer-treatment research is only one potential application. Yang also wants to use Chelatron to develop improved chemical methods for separating plutonium, an activity connected to the laboratory's nuclear materials work.

The current purification process relies on methods developed during the 1940s. Technicians use a chemical called tributyl phosphate, commonly abbreviated TBP, as part of a technique known as solvent extraction. The process separates plutonium from a contaminated mixture by transferring it into another liquid phase and subsequently recovering it in a cleaner solution.

TBP is an organic liquid, meaning it is based on carbon-containing chemical compounds. Its role is to extract plutonium from one solution and carry it into a separate liquid that does not mix with the original. The plutonium can then be released into a cleaner solution as part of the purification sequence.

Although this technique has served the laboratory for approximately 80 years, Yang describes the process as slow and labor-intensive. Her team is now applying SI models to evaluate alternative extractants, chemicals that could perform the separation more efficiently.

The research has a direct connection to nuclear weapons production. Purified plutonium is used in manufacturing plutonium pits, the central cores of nuclear weapons. A faster purification process could therefore accelerate work involving those components. Yang predicts the researchers will identify an alternative that performs better than TBP, although the source does not report that such a replacement has already been demonstrated.

Los Alamos and NVIDIA collaboration

The project combines different areas of expertise. Los Alamos contributes its capabilities in quantum chemistry, scientific simulation and nuclear research. NVIDIA brings high-performance computing and SI foundation models, which provide the computational infrastructure and modeling capabilities used in the discovery workflow.

Mark Chadwick, Los Alamos associate laboratory director for Simulation, Computing and Theory, characterized the initial progress as encouraging. He highlighted the rapid collaboration between the two organizations and the opportunity for scientific expertise and computational methods to reinforce each other.

The partnership illustrates how specialized scientific knowledge can be combined with computational models to investigate problems that are difficult to solve through traditional research alone. Rather than replacing chemists, Chelatron is intended to give them a broader set of possibilities and a more efficient way to decide which ones deserve experimental attention.

Why it matters

Chelatron addresses a bottleneck that reaches across medicine, industrial chemistry and nuclear materials research: the difficulty of designing molecules that selectively recognize one element among several chemically similar alternatives.

Its initial results offer a measurable demonstration of the approach. The system explored tens of thousands of molecular possibilities, produced hundreds of promising computational candidates and reduced the field to five compounds for laboratory validation. That is a substantial change in how researchers can organize the early stages of molecular discovery, although the compounds' real-world performance has yet to be established.

For cancer treatment, the immediate question is whether the five selected candidates can securely bind actinium-225 and support the precise delivery required by targeted alpha therapy. For plutonium processing, researchers must determine whether new extractants can outperform a purification method that has been used for decades.

The next milestones are therefore experimental rather than computational. Laboratory synthesis and testing will show whether Chelatron's predictions translate into useful chemical performance. The results will also help establish whether its iterative design process can consistently identify molecules that conventional research methods might miss.

Yang's Chihuahua-and-muffin comparison captures the underlying scientific challenge: two substances can look remarkably similar while still containing differences that matter enormously. Chelatron is designed to find those differences and turn them into practical molecular designs. Its significance will depend on what happens when those designs leave the computer and enter the laboratory.

Read the original at Los Alamos Reporter

More news

Written by DomainHunt from reporting by Los Alamos Reporter. Read the original article for the full story. Not financial advice.