Advancing Materials for Energy and Sustainability through Computational Discovery
Randall Snurr and his research group develop computational and machine-learning tools to discover, screen, and design nanoporous materials for carbon capture, hydrogen storage, and other applications.
Randall Snurr leads a research group at Northwestern University focused on developing new materials to address important problems in energy and sustainability. His team has made important contributions to the development of nanoporous materials such as metal-organic frameworks (MOFs) for applications including hydrogen storage and carbon capture.
Along the way, Snurr, the John G. Searle Professor of Chemical and Biological Engineering, confronted a problem that has only become more pressing with the rise of AI: the design space for metal-organic frameworks (MOFs) has become far too large to explore experimentally. Because MOFs can be assembled from countless combinations of metal nodes, organic linkers, and topologies, researchers are no longer limited by a lack of ideas. Instead, they face the challenge of sorting through massive numbers of possibilities to determine which materials are worth making.
Establishing computational approaches to navigate an enormous design space
To accelerate discovery in this enormous design space, Snurr and his collaborators have applied increasingly advanced computational tools, moving from large-scale virtual screening to machine learning and generative AI. In the process, they helped establish many of the computational approaches that now underpin AI-driven materials discovery of nanoporous materials.
One of the earliest milestones came in 2012, with what was probably the largest computational study of MOFs ever performed at the time. Snurr's team screened more than 130,000 hypothetical frameworks to identify promising materials for separating carbon dioxide from power-plant exhaust and natural-gas streams. More importantly, the study revealed clear structure-property relationships that had been impossible to see in smaller datasets, giving researchers practical design rules for tailoring pore size, surface area, and chemistry to specific gas separation applications.
Creating new machine learning tools to screen promising materials
Even large-scale screening had its limits. Simulating the adsorption properties of every candidate remained computationally expensive, making it difficult to search the rapidly growing number of known and hypothetical MOFs. To tackle this problem, Snurr's group began incorporating machine learning to accelerate that process.
In 2019, the team developed an interpretable machine-learning model that predicted hydrogen storage performance from energy-based descriptors derived from a material's adsorption landscape. The approach was more than three orders of magnitude faster than conventional molecular simulations while still identifying top-performing candidates, including one that was later validated experimentally. Rather than completely replacing physics-based simulations, the model focused the simulations on the most promising materials, dramatically increasing the speed of computational screening.
Building a generative AI platform capable of designing new materials
Screening existing databases is ultimately limited by the materials they contain. In 2021, Snurr's group helped take the next step by developing a generative AI platform capable of designing entirely new MOFs for carbon capture. Using a supramolecular variational autoencoder, the model learned the underlying relationships between MOF structures and their properties, then generated new framework candidates optimized for gas separation. Many of the AI-designed materials proved competitive with the best-performing MOFs reported in the literature.
By 2024, many of the approaches Snurr had helped pioneer had become central to computational materials discovery. High-throughput screening, machine learning, and generative AI were no longer separate techniques but rather complementary tools for navigating vast chemical design spaces. In a Nature Energy review, Snurr and his coauthors traced that evolution, highlighting how computational screening had progressed from evaluating existing materials to reliably predicting, and in some cases discovering, new ones before they were synthesized in the laboratory.
That progression mirrors the broader trajectory of AI in scientific discovery. Early computational methods accelerated the evaluation of known materials. Machine learning made those searches dramatically more efficient. Generative AI now allows researchers to propose entirely new candidates tailored to specific applications. Across each of those transitions, Snurr's work helped define how AI could move beyond analyzing materials to actively guiding their discovery.
Selected Publications
Wilmer, C. E., Farha, O. K., Bae, Y.-S., Hupp, J. T., & Snurr, R. Q. (2012). Structure–property relationships of porous materials for carbon dioxide separation and capture. Energy & Environmental Science, 5(12), 9849–9856. https://doi.org/10.1039/C2EE23201D
Bucior, B. J., Bobbitt, N. S., Islamoglu, T., Goswami, S., Gopalan, A., Yildirim, T., Farha, O. K., Bagheri, N., & Snurr, R. Q. (2019). Energy-based descriptors to rapidly predict hydrogen storage in metal–organic frameworks. Molecular Systems Design & Engineering, 4(1), 162–174. https://doi.org/10.1039/C8ME00050F
Yao, Z., Sánchez-Lengeling, B., Bobbitt, N. S., Bucior, B. J., Kumar, S. G. H., Collins, S. P., Burns, T., Woo, T. K., Farha, O. K., Snurr, R. Q., & Aspuru-Guzik, A. (2021). Inverse design of nanoporous crystalline reticular materials with deep generative models. Nature Machine Intelligence, 3, 76–86. https://doi.org/10.1038/s42256-020-00271-1
Moghadam, P. Z., Chung, Y. G., & Snurr, R. Q. (2024). Progress toward the computational discovery of new metal–organic framework adsorbents for energy applications. Nature Energy, 9, 121–133. https://doi.org/10.1038/s41560-023-01417-2