Aug 2026· Academia Catalysis· Vol 2· 0 citations· 175 references
TL;DR
It is shown that explainable artificial intelligence (XAI) provides a transformative approach by connecting model predictions to interpreting chemical descriptors, structural motifs, and reaction features, and it is reshaping data-driven catalysis from opaque prediction toward mechanism-informed catalyst design.
Abstract
Catalysis is fundamental to chemical manufacturing, energy conversion, and environmental remediation, yet the rational design of high-performance catalysts remains constrained by the complexity of catalysts, reaction environments, and multi-step reaction networks. However, while machine learning and deep learning have accelerated catalyst discovery and performance prediction, they often obscure the underlying physicochemical mechanisms, limiting mechanistic insights and practical reliability. Here, we show that explainable artificial intelligence (XAI) provides a transformative approach by connecting model predictions to interpre1 chemical descriptors, structural motifs, and reaction features. This review summarizes recent advances in XAI for catalysis, covering key methodologies, evaluation criteria, and representative applications across heterogeneous, homogeneous, and enzymatic catalysis. Particular attention is given to the identification of electronic, structural, and intrinsic atomic descriptors, as well as to the interpretation of active sites, dynamic speciation, reaction pathways, thermodynamics, and kinetics. By transforming black-box predictions into interpretable chemical insights, XAI is reshaping data-driven catalysis from opaque prediction toward mechanism-informed catalyst design.
The electrochemical nitrogen reduction reaction (NRR) offers a promising route for sustainable ammonia synthesis under ambient conditions. Yet, its practical development remains constrained by low catalytic activity, poor selectivity against the competing hydrogen evolution reaction (HER), and the limited availability of rigorously validated experimental data. In recent years, machine learning (ML) has emerged as a powerful tool for accelerating NRR catalyst discovery by mining heterogeneous literature datasets, revealing structure–property relationships, and identifying unexplored catalyst electrolyte operating windows. This review presents a comprehensive account of ML applications in NRR, covering curated experimental databases, feature engineering based on atomic, structural, and DFT‐derived descriptors. Particular emphasis is placed on ML‐guided insights into single‐atom, dual‐atom, alloy, oxide, nitride, and defect‐engineered catalysts. Descriptor‐driven design rules are used to identify promising catalyst motifs through DFT screening and gradient‐boosted regression. The review critically examines the limitations of current ML‐assisted NRR research, including data scarcity, and the mechanistic overlap between NRR and HER. In this context, uncertainty quantification and interpretability tools such as SHAP analysis and partial dependence plots are highlighted as valuable strategies for improving model reliability. Finally, emerging closed‐loop DFT‐ML‐experiment workflows, active learning, and ML models are discussed as essential pathways toward practical, scalable ammonia electrosynthesis.
Oxygen reduction and evolution reactions (ORR and OER) are key electrochemical processes central to energy conversion and chemical transformation. However, the inherently complex, multi-physics nature of ORR/OER-together with diverse operating environments-poses significant challenges to the rational design of electrocatalysts based on structure-property relationships. To overcome these challenges, we developed Two-Stage Material Screening (TSMS), an AI-driven framework that integrates density functional theory (DFT) computations, an active-learning-guided experimental feedback loop, and mechanistic interpretation to enable rapid discovery and systematic evaluation of promising electrocatalysts. Demonstrated in protonic solid oxide cells (P-SOCs), TSMS screened 6,940,032 compositions and identified top-performing candidates that were experimentally validated, achieving a peak power density of 2.68 W cm-2 in fuel cell mode and a current density of 3.51 A cm-2 at 1.3 V in electrolysis mode, with stable performance maintained over 500 h at 600°C. Our analysis revealed that electron affinity is strongly associated with thermodynamic stability, while d-p hybridization and densification resistance emerge as the primary descriptors governing electrocatalytic activity. By combining predictive modeling with mechanistic understanding, TSMS establishes a versatile and broadly generalizable paradigm for accelerating materials discovery.
Xueyu Hu, Yucun Zhou, Haoyu Li et al.· Advances in Materials· 0 citations
Emerging opportunities in physics-informed machine learning, graph neural networks, generative artificial intelligence, active learning, and autonomous closed-loop DFT-ML-MKM workflows are discussed as promising directions for accelerating the discovery of next-generation electrocatalysts with enhanced activity, selectivity, and long-term stability.
Swetarekha Ram, Shalini Tomar, S. Bhattacharjee· Chemical Communications· 0 citations
Catalysts are essential for sustainable chemical manufacturing, yet discovering novel architectures remains a bottleneck dominated by trial-and-error experimentation and computationally intensive screening. In complex reactions such as electrochemical carbon dioxide reduction, product selectivity is governed by dynamic interfacial, electrolyte, and potential factors as well as kinetic pathway competition. Conventional descriptor-based machine learning and computational potentials struggle to resolve these mechanistic branch points, primarily relying on static ground-state descriptors or bulk structural correlations rather than end-to-end topological pathway analysis. Here, we show that frontier language models, when strictly constrained to reason over explicit reaction networks, can discover novel catalysts by identifying the physical levers that govern pathway competition. We developed a human-AI co-thinking framework that enforces network invariance to extract testable hypotheses from complex chemical graphs. Applied to CO2 electroreduction, the framework identified ketene desorption and hydroxide capture as the acetate-forming pathway, and predicted a distinct adsorbed CO and CH2 coupling route to ketene. By isolating actionable control levers, specifically local alkalinity, controlled iron incorporation, and restricted interfacial proton-donor accessibility, the framework guided the prospective synthesis of a copper-iron oxide catalyst demonstrating a threefold increase in acetate selectivity over matched Cu-rich baselines. This mechanism-guided reasoning architecture shifts the computational paradigm from retrospective statistical prediction to forward-looking hypothesis generation, providing a broadly applicable blueprint for mechanism-guided materials discovery.
Sutanay Choudhury, Anwesha Banerjee, Udishnu Sanyal et al.· 0 citations
The integration of artificial intelligence and machine learning into chemical engineering represents a paradigm transformation that fundamentally reconceptualizes practice across molecular, process, and industrial scales. This comprehensive narrative review synthesizes recent advances in AI/ML methodologies, including graph neural networks, physics-informed neural networks, deep reinforcement learning, and generative artificial intelligence – and critically evaluates their applications spanning molecular property prediction, catalyst design, pharmaceutical development, reactor optimization, process control, and sustainability initiatives. Contemporary AI/ML approaches demonstrate unprecedented capabilities in navigating complex multiscale phenomena while maintaining computational tractability and physical interpretability. Landmark achievements include 71% reduction in experimental iterations for reaction optimization through deep reinforcement learning, 98% accuracy in predictive maintenance using LSTM-based fault detection, sub-1% prediction errors in virtual metrology for semiconductor manufacturing, and substantial improvements incarbon capture efficiency through machine learning-guided materials discovery. Physics-informed neural networks address the critical challenge of plant-model mismatch by synergistically integrating mechanistic knowledge with data-driven learning, enabling extrapolation beyond training domains while respecting conservation laws. Explainable AI techniques, particularly SHAP analysis, enhance operational acceptance by 52% in safety-critical applications through transparent decision-making pathways. Despite remarkable progress, persistent challenges remain in data quality and standardization, model interpretability for regulatory compliance, computational scalability for real-time control, and integration with legacy industrial infrastructure. The review identifies transformative future directions including multi-modal learning frameworks, transfer learning for data-scarce applications, quantum machine learning for molecular design, and human-AI collaborative systems. Successful deployment demands interdisciplinary collaboration uniting chemical engineering domain expertise with computational intelligence, guided by principles of transparency, reproducibility, and responsible innovation to address sustainability imperatives while maintaining operational excellence and safety in chemical manufacturing.
Fatemeh Valizadeh Hajidehi, Azam Mina· Chemical and Process Enginee...· 0 citations
The discovery and development of high-performance catalysts, which is crucial across all catalysis areas, requires advanced technologies and innovative approaches. Recently, machine learning (ML) has shown promise in accelerating this process, but its capability and examples of discovery of truly novel catalysts have remained limited. In this study, we describe an ML approach that goes beyond the traditional element pool, incorporating elements that have not been previously studied, to develop highly efficient catalysts for ethanol synthesis via CO2 hydrogenation. Starting with an initial data set of 58 catalysts (274 data points obtained at reaction temperatures ranging from 240-400 °C), we conducted 24 iterations of a closed-loop discovery system (ML predictions + experimental validation), testing a total of 555 catalysts (2477 data points), and building a large experimental data set. More than 50 catalysts with superior activity were discovered through this data-driven approach. The multielemental Pd(0.8)-Au(0.3)/K(2.5)-Sr(1)-Fe(20)-Zn(4)-Cd(2)-Yb(1)-Re(1)/CeO2(25%)-ZrO2 catalyst, where the numbers in parentheses represent weight percent (wt %), was identified as the most effective catalyst for ethanol synthesis (ethanol space-time yield: 8.2 mmol gcat-1 h-1 with a CO2 conversion of 57.6% and an ethanol selectivity of 23.2% under reaction conditions of 360 °C, 4 MPa, 12 L gcat-1 h-1, H2/CO2 = 3/1). Comprehensive characterizations, including in situ/operando techniques such as X-ray absorption spectroscopy (XAS), ambient-pressure X-ray photoelectron spectroscopy (AP-XPS), and diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), enable us to highlight the critical roles of each constituting element in improving ethanol synthesis efficiency.
Pengfei Du, A. Ait El Fakir, S. Mine et al.· Journal of the American Chem...· 0 citations