CatPath-GPT: A Mixture of Experts System for Computational Catalyst Design

Abstract

Despite advances in computational catalysis, the complexity of theoretical calculations and specialised expertise requirements limit the broader adoption of catalyst design tools. This work introduces CatPath-GPT, a mixture-of-experts framework that democratizes computational catalyst design by integrating three AI specialists: product prediction (77.2% accuracy), computational planning, and automated code generation through a unified BERT-based router. Experimental validation through two case studies demonstrates practical impact: systematic screening of Cu_xZn_1−x catalysts identifies optimal compositions for selective CO₂RR (Cu₇₅Zn₂₅ for ethanol), while high-throughput metal oxide screening reproduces Nørskov's classical scaling relationships and identifies high-activity materials. Benchmarking against GPT-4 and Mistral-7B demonstrates superior performance across catalyst-related tasks, particularly in modeling complex surface reactions. The open-source framework enables researchers without computational expertise to perform advanced catalyst design, potentially transforming how catalytic materials are discovered and optimized.

Keywords

catalyst design / CO₂ reduction reaction / large language models / mixture of experts / reaction mechanism

References

[1] Y. Xie, P. Ou, X. Wang, et al., “High Carbon Utilization in CO₂ Reduction to Multi-Carbon Products in Acidic Media,” Nature Catalysis 5, no. 6 (2022): 564–570.

[2] W. Luc, X. Fu, J. Shi, et al., “Two-Dimensional Copper Nanosheets for Electrochemical Reduction of Carbon Monoxide to Acetate,” Nature Catalysis 2, no. 5 (2019): 423–430.

[3] T. T. Zhuang, Z. Q. Liang, A. Seifitokaldani, et al., “Steering Post-C–C Coupling Selectivity Enables High Efficiency Electroreduction of Carbon Dioxide to Multi-Carbon Alcohols,” Nature Catalysis 1, no. 6 (2018): 421–428.

[4] M. Jouny, W. Luc, and F. Jiao, “High-Rate Electroreduction of Carbon Monoxide to Multi-Carbon Products,” Nature Catalysis 1, no. 10 (2018): 748–755.

[5] K. Jiang, R. B. Sandberg, A. J. Akey, et al., “Metal Ion Cycling of Cu Foil for Selective C–C Coupling in Electrochemical CO₂ Reduction,” Nature Catalysis 1, no. 2 (2018): 111–119.

[6] C. T. Dinh, T. Burdyny, M. G. Kibria, et al., “CO₂ Electroreduction to Ethylene via Hydroxide-Mediated Copper Catalysis at an Abrupt Interface,” Science 360, no. 6390 (2018): 783–787.

[7] Q. Lu, J. Rosen, Y. Zhou, et al., “A Selective and Efficient Electrocatalyst for Carbon Dioxide Reduction,” Nature Communications 5, no. 1 (2014): 3242.

[8] K. P. Kuhl, T. Hatsukade, E. R. Cave, D. N. Abram, J. Kibsgaard, and T. F. Jaramillo, “Electrocatalytic Conversion of Carbon Dioxide to Methane and Methanol on Transition Metal Surfaces,” Journal of the American Chemical Society 136, no. 40 (2014): 14107–14113.

[9] S. Chu and A. Majumdar, “Opportunities and Challenges for a Sustainable Energy Future,” Nature 488, no. 7411 (2012): 294–303.

[10] J. P. Gattuso, A. Magnan, R. Billé, et al., “Contrasting Futures for Ocean and Society From Different Anthropogenic CO₂ emissions Scenarios,” Science 349, no. 6243 (2015): aac4722.

[11] J. B. Greenblatt, D. J. Miller, J. W. Ager, F. A. Houle, and I. D. Sharp, “The Technical and Energetic Challenges of Separating (Photo)Electrochemical Carbon Dioxide Reduction Products,” Joule 2, no. 3 (2018): 381–420.

[12] Y. Chen, Z. Fan, J. Wang, et al., “Ethylene Selectivity in Electrocatalytic CO₂ Reduction on Cu Nanomaterials: A Crystal Phase-Dependent Study,” Journal of the American Chemical Society 142, no. 29 (2020): 12760–12766.

[13] M. Xie, Y. Shen, W. Ma, et al., “Fast Screening for Copper-Based Bimetallic Electrocatalysts: Efficient Electrocatalytic Reduction of CO₂ to C₂⁺ Products on Magnesium-Modified Copper,” Angewandte Chemie International Edition 61, no. 51 (2022): e202213423.

[14] R. Zhao, Y. Wang, G. Ji, et al., “Partially Nitrided Ni Nanoclusters Achieve Energy-Efficient Electrocatalytic CO₂ Reduction to CO at Ultralow Overpotential,” Advanced Materials 35, no. 5 (2023): 2205262.

[15] J. L. Hitt, Y. C. Li, S. Tao, et al., “A High Throughput Optical Method for Studying Compositional Effects in Electrocatalysts for CO₂ Reduction,” Nature Communications 12 (2021): 1114.

[16] E. Reddington, A. Sapienza, B. Gurau, et al., “Combinatorial Electrochemistry: A Highly Parallel, Optical Screening Method for Discovery of Better Electrocatalysts,” Science 280, no. 5370 (1998): 1735–1737.

[17] S. Ringe, “The Importance of a Charge Transfer Descriptor for Screening Potential CO₂ Reduction Electrocatalysts,” Nature Communications 14 (2023): 2598.

[18] R. A. Hoyt, M. M. Montemore, I. Fampiou, W. Chen, G. Tritsaris, and E. Kaxiras, “Machine Learning Prediction of H Adsorption Energies on Ag Alloys,” Journal of Chemical Information and Modeling 59, no. 4 (2019): 1357–1365.

[19] S. Nayak, S. Bhattacharjee, J. H. Choi, and S. C. Lee, “Machine Learning and Scaling Laws for Prediction of Accurate Adsorption Energy,” Journal of Physical Chemistry A 124, no. 1 (2020): 247–254.

[20] D. H. Mok, H. Li, G. Zhang, C. Lee, K. Jiang, and S. Back, “Data-Driven Discovery of Electrocatalysts for CO₂ Reduction Using Active Motifs-Based Machine Learning,” Nature Communications 14 (2023): 7303.

[21] B. C. Yeo, H. Nam, H. Nam, et al., “High-Throughput Computational–Experimental Screening Protocol for the Discovery of Bimetallic Catalysts,” npj Computational Materials 7 (2021): 1232–1241.

[22] T. Lan, H. Wang, and Q. An, “Enabling High Throughput Deep Reinforcement Learning With First Principles to Investigate Catalytic Reaction Mechanisms,” Nature Communications 15 (2024): 6281.

[23] T. B. Brown, B. Mann, N. Ryder, et al., “Language Models Are Few-Shot Learners,” Advances in Neural Information Processing Systems 33, no. 159 (2020): 1877–1901.

[24] A. Vaswani, N. Shazeer, N. Parmar, et al., “Attention is All You Need,” in Advances in Neural Information Processing Systems 30 (NeurIPS 2017) (Curran Associates, Inc., 2017), 6000–6010.

[25] J. Qiu, K. Lam, G. Li, et al., “LLM-Based Agentic Systems in Medicine and Healthcare,” Nature Machine Intelligence 6, no. 12 (2024): 1418–1420.

[26] D. Peng, L. Zheng, D. Liu, et al., “Large-Language Models Facilitate Discovery of the Molecular Signatures Regulating Sleep and Activity,” Nature Communications 15 (2024): 3685.

[27] X. Liu, H. Liu, G. Yang, et al., “A Generalist Medical Language Model for Disease Diagnosis Assistance,” Nature Medicine 31, no. 3 (2025): 932–942.

[28] A. V. Gougherty and H. L. Clipp, “Testing the Reliability of an AI-Based Large Language Model to Extract Ecological Information From the Scientific Literature,” npj Biodiversity 3 (2024): 13.

[29] Z. Lin, H. Akin, R. Rao, et al., “Evolutionary-Scale Prediction of Atomic-Level Protein Structure With a Language Model,” Science 379, no. 6637 (2023): 1123–1130.

[30] A. Madani, B. Krause, E. R. Greene, et al., “Large Language Models Generate Functional Protein Sequences Across Diverse Families,” Nature Biotechnology 41, no. 8 (2023): 1099–1106.

[31] A. M. Bran, S. Cox, O. Schilter, C. Baldassari, A. D. White, and P. Schwaller, “Augmenting Large Language Models With Chemistry Tools,” Nature Machine Intelligence 6, no. 5 (2024): 525–535.

[32] Y. Kang and J. Kim, “ChatMOF: An Artificial Intelligence System for Predicting and Generating Metal-Organic Frameworks Using Large Language Models,” Nature Communications 15 (2024): 4705.

[33] L. M. Antunes, K. T. Butler, and R. Grau-Crespo, “Crystal Structure Generation With Autoregressive Large Language Modeling,” Nature Communications 15 (2024): 10570.

[34] G. Kresse and J. Furthmüller, “Efficiency of Ab Initio Total Energy Calculations for Metals and Semiconductors Using a Plane-Wave Basis Set,” Computational Materials Science 6, no. 1 (1996): 15–50.

[35] W. Kohn and L. J. Sham, “Self-Consistent Equations Including Exchange and Correlation Effects,” Physical Review 140, no. 4A (1965): A1133–A1138.

[36] W. Fedus, B. Zoph, and N. Shazeer, “Switch Transformers: Scaling to Trillion Parameter Models With Simple and Efficient Sparsity,” Journal of Machine Learning Research 23, no. 1 (2022): 5232–5270.

[37] J. Wei and K. Zou, “EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks,” in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (2019), 6382–6388.

[38] M. Kim, S. Lee, S. J. Hong, D. S. Chang, and J. Choi, “Understanding and Improving Knowledge Distillation for Quantization Aware Training of Large Transformer Encoders,” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP 2022) (Association for Computational Linguistics, 2022), 6713–6725.

[39] J. Devlin, M. W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT 2019) (Association for Computational Linguistics, 2019), 4171–4186.

[40] J. Varghese and J. Chapiro, “ChatGPT: The Transformative Influence of Generative AI on Science and Healthcare,” Journal of Hepatology 80, no. 6 (2024): 977–980.

[41] V. Sanh, L. Debut, J. Chaumond, and T. Wolf, “DistilBERT, A Distilled Version of BERT: Smaller, Faster, Cheaper and Lighter,” preprint arXiv, October 2, 2019, https://arxiv.org/abs/1910.01108v4.

[42] OpenAI, J. Achiam, S. Adler, et al., “GPT-4 Technical Report,” preprint arXiv, March 15, 2023, https://arxiv.org/abs/2303.08774v6.

[43] OpenAI, “Hello GPT-4o,” published May 13, 2024, https://openai.com/index/hello-gpt-4o/.

[44] H. Touvron, L. Martin, K. Stone, et al., “Llama 2: Open Foundation and Fine-Tuned Chat Models,” preprint arXiv, July 18, 2023, https://arxiv.org/abs/2307.09288v2.

[45] A. Q. Jiang, A. Sablayrolles, A. Roux, et al., “Mixtral of Experts,” preprint arXiv, January 8, 2024, https://arxiv.org/abs/2401.04088v1.

[46] A. Q. Jiang, A. Sablayrolles, A. Mensch, et al., “Mistral 7B,” preprint arXiv, October 10, 2023, https://arxiv.org/abs/2310.06825v1.

[47] E. J. Hu, Y. Shen, P. Wallis, et al., “LoRA: Low-Rank Adaptation of Large Language Models,” preprint arXiv, June 17, 2021, https://arxiv.org/abs/2106.09685.

[48] Y. Zhao, J. Huang, J. Hu, et al., “SWIFT: A Scalable Lightweight Infrastructure for Fine-Tuning,” preprint arXiv, August 10, 2024, https://arxiv.org/abs/2408.05517v3.

[49] G. Mischler, Y. A. Li, S. Bickel, A. D. Mehta, and N. Mesgarani, “Contextual Feature Extraction Hierarchies Converge in Large Language Models and the Brain,” Nature Machine Intelligence 6, no. 12 (2024): 1467–1477.

[50] C. Raffel, N. Shazeer, A. Roberts, et al., “Exploring the Limits of Transfer Learning With a Unified Text-to-Text Transformer,” Journal of Machine Learning Research 21, no. 1 (2020): 5485–5551.

[51] Y. Liu, M. Ott, N. Goyal, et al., “RoBERTa: A Robustly Optimized BERT Pretraining Approach,” preprint arXiv, July 26, 2019, https://arxiv.org/abs/1907.11692v1.

[52] S. Liu, Y. Fan, Y. Wang, et al., “Surface-Oxygen-Rich Bi@C Nanoparticles for High-Efficiency Electroreduction of CO₂ to Formate,” Nano Letters 22, no. 22 (2022): 9107–9114.

[53] F. Yang, J. Pei, and H. Zhao, “Insight Understanding of Ultrathin Carbon-Deficient Molybdenum Carbide Catalytic Activity for CO₂ Conversion Into Hydrocarbon Fuels,” Journal of Physical Chemistry C 127, no. 25 (2023): 11978–11985.

[54] L. Vitos, A. V. Ruban, H. L. Skriver, and J. Kollár, “The Surface Energy of Metals,” Surface Science 411, no. 1–2 (1998): 186–202.

[55] S. Okamoto and A. J. Millis, “Theory of Mott Insulator–Band Insulator Heterostructures,” Physical Review B 70 (2004): 075101.

RIGHTS & PERMISSIONS

2026 The Author(s). SmartMat published by Tianjin University and John Wiley & Sons Australia, Ltd.