Reaction-anchored generative informatics for million-scale CO₂ catalyst screening via multi-adsorbate constraints

Abstract

The rational design of electrocatalysts for multi-electron reaction networks hinges on navigating vast, multidimensional surface-environment spaces defined by coupled variations in composition, coordination topology, and adsorbate configurations. Taking electrochemical carbon dioxide reduction to ethanol as a model challenge, traditional screening methodologies struggle to efficiently represent and search these combinatorial landscapes. Here, we present a reaction-constrained generative materials informatics framework anchored by late-stage oxygenated intermediates (*OCH₂CH₃). A fine-tuned generative Transformer model efficiently sampled 1,000,000 adsorption configurations. To efficiently navigate this massive space, adsorption configurations were compressed into latent topological representations via oxygen-centered local node embeddings derived from the universal models for atoms (UMA) platform. Cluster-guided spatial navigation based on these UMA-derived representations achieved a 15-fold enrichment in positive-candidate discovery over naive representative sampling. By jointly applying multi-adsorbate thermodynamic constraints (*OCH₂CH₃, *CO, and *H) within the CO₂RR network, the workflow down-selected 75 promising material candidates. Density functional theory (DFT) calculations benchmarked representative systems, validating the screening-level fidelity of UMA-predicted binding energies. Furthermore, subgroup discovery extracted interpretable, physically intuitive elemental rules governing candidate enrichment. CO-CO coupling calculations further identified Ga-Pt and Al-Cu-Pd as kinetically favored over Cu. This framework successfully recovers 13 reported CO₂RR-related materials while uncovering 62 unexplored compositional domains, providing a transferable paradigm for reaction-constrained generative surface informatics.

Keywords

Materials informatics / generative Transformer / million-scale screening / multi-adsorbate constraints / CO₂ reduction reaction

References

[1] Bushuyev OS,Dinh CT.What should we make with CO₂ and how can we make it?.Joule2018;2:825-32

[2] De Luna, P.; Hahn, C.; Higgins, D.; Jaffer, S. A.; Jaramillo, T. F.; Sargent, E. H. What would it take for renewably powered electrosynthesis to displace petrochemical processes?.Science2019;364:eaav3506

[3] Jouny M,Jiao F.General techno-economic analysis of CO₂ electrolysis systems.Ind Eng Chem Res2018;57:2165-77

[4] Nitopi S,Scott SB.Progress and perspectives of electrochemical CO₂ reduction on copper in aqueous electrolyte.Chem Rev2019;119:7610-72

[5] Zhu P.High-purity and high-concentration liquid fuels through CO₂ electroreduction.Nat Catal2021;4:943-51

[6] Tran K.Active learning across intermetallics to guide discovery of electrocatalysts for CO₂ reduction and H₂ evolution.Nat Catal2018;1:696-703

[7] Zhao F,Zhang Y,Zhang J.A review of CO₂ electroreduction to ethanol: C-C coupling mechanistic insights and catalyst design.Nanomicro Lett2026;18:313 PMCID:PMC13049153

[8] Li F,Wang Z.Cooperative CO₂-to-ethanol conversion via enriched intermediates at molecule–metal catalyst interfaces.Nat Catal2020;3:75-82

[9] Luo M,Li YC.Hydroxide promotes carbon dioxide electroreduction to ethanol on copper via tuning of adsorbed hydrogen.Nat Commun2019;10:5814 PMCID:PMC6925210

[10] Zhan C,Rettenmaier C.Key intermediates and Cu active sites for CO₂ electroreduction to ethylene and ethanol.Nat Energy2024;9:1485-96 PMCID:PMC11659170

[11] Liu X,Wang Z,Deng Y.Building catalyst exploration highways by integrating high‐throughput and machine learning technologies.Adv Energy Mater2026;16:e05497

[12] Levy O,Curtarolo S.Uncovering compounds by synergy of cluster expansion and high-throughput methods.J Am Chem Soc2010;132:4830-3

[13] Greeley J,Bonde J,Nørskov JK.Computational high-throughput screening of electrocatalytic materials for hydrogen evolution.Nat Mater2006;5:909-13

[14] Curtarolo S,Nardelli MB,Sanvito S.The high-throughput highway to computational materials design.Nat Mater2013;12:191-201

[15] Pan Y,Cai F,Xu J.Accelerating the discovery of oxygen reduction electrocatalysts: high-throughput screening of element combinations in Pt-based high-entropy alloys.Angew Chem Int Ed Engl2024;63:e202407116

[16] Ivanciuc O.Applications of support vector machines in chemistry. In Reviews in computational chemistry, Vol 23; Wiley, 2007; pp. 291-400.

[17] Machado Cavalcanti, F.; Emilia Kozonoe, C.; André Pacheco, K.; Maria de Brito Alves, R. Application of artificial neural networks to chemical and process engineering. In Deep learning applications. IntechOpen; 2021.

[18] Myles AJ,Liu Y,Brown SD.An introduction to decision tree modeling.J Chemom2004;18:275-85

[19] Butler KT,Cartwright H,Walsh A.Machine learning for molecular and materials science.Nature2018;559:547-55

[20] Ahneman DT,Lin S,Doyle AG.Predicting reaction performance in C-N cross-coupling using machine learning.Science2018;360:186-90

[21] Reid JP.Holistic prediction of enantioselectivity in asymmetric catalysis.Nature2019;571:343-8 PMCID:PMC6641578

[22] Huang B.Ab initio machine learning in chemical compound space.Chem Rev2021;121:10001-36 PMCID:PMC8391942

[23] Han Z,Wang T.Machine-learning-assisted design of a binary descriptor to decipher electronic and structural effects on sulfur reduction kinetics.Nat Catal2023;6:1073-86

[24] Esterhuizen JA,Linic S.Interpretable machine learning for knowledge generation in heterogeneous catalysis.Nat Catal2022;5:175-84

[25] Song Z,Liu F.Distilling universal activity descriptors for perovskite catalysts from multiple data sources via multi-task symbolic regression.Mater Horiz2023;10:1651-60

[26] Weng B,Zhu R.Simple descriptor derived from symbolic regression accelerating the discovery of new perovskite catalysts.Nat Commun2020;11:3513 PMCID:PMC7360597

[27] Zhong M,Min Y.Accelerated discovery of CO₂ electrocatalysts using active machine learning.Nature2020;581:178-83

[28] Ulissi ZW,Xiao J.Machine-learning methods enable exhaustive searches for active bimetallic facets and reveal active site motifs for CO₂ reduction.ACS Catal2017;7:6600-8

[29] Sun J,Xu Y.Machine learning aided design of single-atom alloy catalysts for methane cracking.Nat Commun2024;15:6036 PMCID:PMC11255339

[30] Zhang Q,Liu X.Multiparameter machine learning quantifies electronic dominance in Pd-catalyzed formic acid dehydrogenation.Nano Lett2026;26:7927-36

[31] Bertheussen E,Ravasio D.Acetaldehyde as an intermediate in the electroreduction of carbon monoxide to ethanol on oxide-derived copper.Angew Chem Int Ed Engl2016;55:1450-4 PMCID:PMC4737304

[32] Piqué O,Handoko AD,Calle-Vallejo F.Selectivity map for the late stages of CO and CO₂ reduction to C₂ species on copper electrodes.Angew Chem Int Ed Engl2021;60:10784-90

[33] Vasileff A,Zhi X.Electrochemical reduction of CO₂ to ethane through stabilization of an ethoxy intermediate.Angew Chem Int Ed Engl2020;59:19649-53

[34] Goodfellow ,Pouget-Abadie ,Mirza M.Generative adversarial nets. In Proceedings of the 28th International Conference on Neural Information Processing Systems, Montréal, Canada, December 8-13, 2014; Ghahramani, Z.; Welling, M.; Cortes, C.; Lawrence, N. D.; Weinberger, K. Q., Eds.; Curran Associates, Inc.: Red Hook, USA, 2014; Vol. 2, pp 2672-80.

[35] Yan D,Chen CC.Structure prediction and materials design with generative neural networks.Nat Comput Sci2023;3:572-4

[36] Zhao Y,Hu M.High-throughput discovery of novel cubic crystal materials using deep generative neural networks.Adv Sci2021;8:e2100566 PMCID:PMC8529451

[37] Yao Z,Bobbitt NS.Inverse design of nanoporous crystalline reticular materials with deep generative models.Nat Mach Intell2021;3:76-86

[38] Zeni C,Zügner D.A generative model for inorganic materials design.Nature2025;639:624-32 PMCID:PMC11922738

[39] Bran AM,Schilter O,White AD.Augmenting large language models with chemistry tools.Nat Mach Intell2024;6:525-35 PMCID:PMC11116106

[40] Jablonka KM,Ortega-Guerrero A.Leveraging large language models for predictive chemistry.Nat Mach Intell2024;6:161-9

[41] Boiko DA,Kline B.Autonomous chemical research with large language models.Nature2023;624:570-8 PMCID:PMC10733136

[42] Antunes LM,Grau-Crespo R.Crystal structure generation with autoregressive large language modeling.Nat Commun2024;15:10570 PMCID:PMC11624194

[43] Mok DH.Generative pretrained transformer for heterogeneous catalysts.J Am Chem Soc2024;146:33712-22

[44] Song Z,Lu S,Zhou Q.Inverse design of promising electrocatalysts for CO₂ reduction via generative models and bird swarm algorithm.Nat Commun2025;16:1053 PMCID:PMC11770065

[45] Li R,Tang Q.Generative intelligence explores the chemical space of ten million catalysts.Chem Sci2026;17:12996-3006 PMCID:PMC13195687

[46] Chanussot L,Goyal S.Open catalyst 2020 (OC20) dataset and community challenges.ACS Catal2021;11:6059-72

[47] Wood BM,Fu X. UMA: a family of universal models for atoms. arXiv 2025, arXiv:2506.23971. Available online: https://doi.org/10.48550/arXiv.2506.23971. (accessed 2026-09-23)

[48] Kresse G.Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set.Comput Mater Sci1996;6:15-50

[49] Kresse G.Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set.Phys Rev B Condens Matter1996;54:11169-86

[50] Hammer B,Nørskov JK.Improved adsorption energetics within density-functional theory using revised Perdew-Burke-Ernzerhof functionals.Phys Rev B1999;59:7413-21

[51] Kresse G.From ultrasoft pseudopotentials to the projector augmented-wave method.Phys Rev B1999;59:1758-75

[52] Atzmueller M.Subgroup discovery.WIREs Data Min Knowl Discov2015;5:35-49

[53] Lemmerich F.pysubgroup: Easy-to-use subgroup discovery in Python. In Machine learning and knowledge discovery in databases. Cham: Springer International Publishing, 2019; Vol. 11053, pp. 658-62.

[54] Shapley LS.17. A VALUE for n-person games. In Contributions to the theory of games (AM-28), Volume II; Princeton University Press, 1953; pp. 307-18.

[55] Mcinnes L,Saul N.UMAP: uniform manifold approximation and projection.J Open Source Softw2018;3:861

[56] Peng C,Luo G.(111) Facet-oriented Cu₂Mg intermetallic compound with Cu₃-Mg sites for CO₂ electroreduction to ethanol with industrial current density.Angew Chem Int Ed Engl2024;63:e202316907

[57] Zhang L,Wu L.Oxophilicity-controlled CO₂ electroreduction to C₂₊ alcohols over Lewis acid metal-doped Cu^δ+ catalysts.J Am Chem Soc2023;145:21945-54

[58] Hou Y,Wang Y.Steering acidic CO₂ electroreduction to multicarbon alcohols with high efficiency and selectivity over calcium-induced bicrystalline Cu architecture.ACS Catal2025;15:19227-37

[59] Zhu Y,Li H.Confinement effect and hydrogen species modulation toward enhanced electrochemical CO₂ reduction to ethanol.Research2025;8:0796 PMCID:PMC12311302

[60] Hoang TTH,Ma S.Nanoporous copper-silver alloys by additive-controlled electrodeposition for the selective electroreduction of CO₂ to ethylene and ethanol.J Am Chem Soc2018;140:5791-7

[61] Kuang S,Li M.Asymmetrical electrohydrogenation of CO₂ to ethanol with copper-gold heterojunctions.Proc Natl Acad Sci U S A2023;120:e2214175120 PMCID:PMC9942890

[62] Zhou X,Zhang J,Zhang X.Electronic state modulation of Cu by Ge for post C–C steps in electrochemical CO₂ reduction.Adv Funct Mater2026;36:e75284

[63] Bae S,Gwon Y,Sohn Y.Interface engineering of Pt-deposited Cu electrodes via laser ablation for enhanced electrochemical CO₂ reduction to multi-carbon products.Adv Ind Eng Chem2025;1:19

[64] Varandili SB,Vavra J.Elucidating the structure-dependent selectivity of CuZn towards methane and ethanol in CO₂ electroreduction using tailored Cu/ZnO precatalysts.Chem Sci2021;12:14484-93 PMCID:PMC8580038

[65] Guo C,Shi Y.Electrocatalytic reduction of CO₂ to ethanol at close to theoretical potential via engineering abundant electron-donating Cu^δ+ species.Angew Chem Int Ed Engl2022;61:e202205909

[66] Crandall BS,Foucher AC.Cu based dilute alloys for tuning the C₂₊ selectivity of electrochemical CO₂ reduction.Small2024;20:e2401656

[67] Li J,Wan M.Silica-copper catalyst interfaces enable carbon-carbon coupling towards ethylene electrosynthesis.Nat Commun2021;12:2808 PMCID:PMC8121866

[68] Guzmán H,Irabien A,Hernández S.Role of electrochemical cell configuration on the selectivity of CuZnAl-oxide-based electrodes for the continuous CO₂ conversion: aqueous electrolyte vs. catholyte-less configuration.Discov Chem Eng2024;4:49