AI-Guided Materials Discovery Produces New Methane-Capture Frameworks
A new machine-learning-guided workflow has produced two promising materials for separating methane from nitrogen, offering a potential way to move climate-related materials research more quickly from academic theory to industrial application. Developed by researchers at the University of Chicago’s Pritzker School of Molecular Engineering and Department of Chemistry, the process links literature mining, computational modeling, materials design, laboratory synthesis and experimental testing within one continuous discovery cycle. The resulting materials, named UCHI-1 and UCHI-2, are zinc-based metal-organic frameworks, or MOFs, designed to capture methane under ambient conditions. The work addresses a persistent problem in materials science: many computationally promising structures remain trapped in databases, unpublished files or dissertations because no experimental team ever synthesizes or tests them.
MOFs are porous crystalline materials built from metal ions or metal clusters connected by organic molecules. Their structures contain networks of precisely defined cavities and channels that can selectively adsorb gases. By changing the metal centers and organic linkers, researchers can tune a MOF’s pore size, internal surface chemistry and affinity for specific molecules. Those properties make MOFs attractive for gas separation, carbon capture, chemical purification and energy applications. In the new study, the researchers focused on the difficult separation of methane from nitrogen. The two gases can occur together in natural gas streams and other industrial environments, yet they have similar physical properties, making selective separation technically demanding and potentially expensive.
Methane received particular attention because its climate impact is disproportionately large. Although methane remains in the atmosphere for roughly a decade, it absorbs infrared radiation far more efficiently than carbon dioxide during that period. Over a 20-year time frame, methane’s warming impact is estimated to be about 80 times greater than that of carbon dioxide. Major sources include livestock and other agricultural activities, landfills, coal mining, oil and natural gas production, and leaks from pipelines and compression equipment. Capturing methane before it escapes could therefore reduce near-term warming while also preserving a valuable fuel and industrial feedstock. The researchers estimate that methane losses from industrial distribution systems cost the sector approximately $10 billion each year.
The study was led by Andrea Darù, a postdoctoral researcher in the Laura Gagliardi Group, with collaboration from Gagliardi, University of Chicago chemistry professor John Anderson and Anderson Lab postdoctoral scholar Jianheng “Allen” Ling. The project was conducted through the Center for Advanced Materials for Environmental Solutions, or CAMES, which Gagliardi co-directs. Rather than treating computation and laboratory experimentation as separate stages, the team designed a workflow intended to keep both sides connected from the beginning. Computational scientists mined information from academic publications and existing materials datasets, then used machine-learning models to identify and refine candidate structures. Experimental researchers subsequently synthesized the most promising candidates and measured their gas-adsorption and separation properties.
This approach is important because conventional materials discovery often breaks down between prediction and reality. A computational group may generate thousands or millions of hypothetical structures, but the files can be difficult for experimentalists to interpret, reproduce or prioritize. Experimental researchers, meanwhile, frequently rely on chemical intuition and trial-and-error methods based on materials that have already been synthesized. Both strategies can produce breakthroughs, but the gap between them means that many potentially useful materials are never made. The new workflow attempts to close that gap by incorporating practical constraints, including whether a material can be synthesized using accessible chemicals, whether its performance can be measured reliably and whether its composition makes sense for eventual manufacturing.
Machine learning served as a guide rather than a replacement for chemical reasoning. The team trained models on data extracted from the scientific literature, allowing the system to learn relationships between molecular structure and gas-adsorption behavior. Candidate MOFs could then be ranked according to predicted methane uptake, methane-to-nitrogen selectivity and other characteristics relevant to industrial separation. The researchers iterated between computational predictions and experimental feedback, using laboratory results to improve the design process. This type of closed-loop strategy can reduce the number of materials that must be synthesized blindly, concentrating time and resources on structures with a stronger probability of working.
The researchers also considered cost while designing the materials. Many MOFs developed for gas capture use metals such as nickel or copper, which can offer desirable chemical and structural properties but may increase the expense of large-scale production. UCHI-1 and UCHI-2 instead use zinc, a comparatively accessible metal. The goal was not simply to maximize separation performance under ideal laboratory conditions, but to identify materials that could provide useful methane-nitrogen separation without relying on unnecessarily expensive components. According to Darù, the new materials achieved slightly better separation than some existing examples while using a less costly metal, a result that could become more significant if the materials prove stable and scalable under industrial operating conditions.
The two MOFs are being presented as proof of concept for the workflow rather than as finished commercial products. Their importance lies both in their measured performance and in the route used to discover them. A material intended for industrial gas separation must satisfy several requirements at once: it must adsorb the target gas effectively, distinguish it from competing molecules, remain stable through repeated cycles, tolerate moisture and contaminants, and be manufactured in sufficient quantities without excessive cost or energy use. It must also be formed into pellets, membranes or other practical configurations that allow gas to flow through a processing system. The new study brings these considerations into the discovery process earlier than is typical, helping ensure that computationally attractive structures are not disconnected from real-world constraints.
The work reflects a broader shift in materials science toward autonomous and semi-automated discovery systems. As machine-learning models become better at recognizing patterns in chemical and structural data, they can help researchers explore areas of design space that would be difficult to investigate manually. Yet the researchers emphasize that algorithms alone cannot solve the materials-development bottleneck. Predictions must be tested, failed designs must inform later decisions, and synthetic chemistry must remain central to the process. By connecting researchers at the University of Chicago with collaborators at Argonne National Laboratory, Northwestern University and industry, the team aims to create a repeatable pathway in which promising materials continue moving forward instead of ending with a publication or a digital structure file.
The methane-separation project ultimately points to a different model for climate technology: one in which discovery is measured not only by how novel a material appears, but also by whether it can survive the journey from computer screen to factory. UCHI-1 and UCHI-2 demonstrate that a single integrated workflow can produce, synthesize and evaluate new MOFs while accounting for performance and cost. The researchers now hope to improve the system and use it to identify materials with even stronger methane-capture capabilities. If successful, the strategy could be applied beyond methane to carbon dioxide removal, gas purification and other environmental challenges, accelerating the movement of advanced materials from academic laboratories into technologies capable of reducing pollution at scale.
Subject of Research: Machine-learning-guided discovery of zinc-based metal-organic frameworks for methane-nitrogen separation
Article Title: End-To-End Discovery of MOFs for Ambient CH4 Adsorption
News Publication Date: 28-Jul-2026
Web References: Journal of the American Chemical Society article; Center for Advanced Materials for Environmental Solutions
References: DOI: 10.1021/jacs.6c07479
Image Credits: UChicago Pritzker School of Molecular Engineering
Keywords
Machine learning, metal-organic frameworks, methane capture, methane-nitrogen separation, climate change, materials science, gas adsorption, sustainable technology, computational chemistry, environmental solutions
Tags: accelerating climate solutions with AIAI-guided climate solution developmentambient-condition methane capture methodsclimate change mitigation through materials sciencecomputational modeling of porous materialsdata-driven materials designexperimental synthesis of methane adsorbentsindustrial application of climate-focused materialsliterature mining in materials researchMachine learning for materials discoverymetal-organic frameworks (MOFs) for gas separationmethane capture technology
