ExplainabilityTransparencyAccountabilityphysics.chem-ph
Beyond the Black Box: An Interpretable Machine Learning Framework for Predicting Electronic Structure Microdescriptors and Structure-Performance Relationships in Fe-based Catalytic Systems
Oyinkansola Romiluyi
May 9, 2026
arXiv.org
Key Takeaways
-
Interpretability enables scientific discovery: By using SHAP-based explainable AI techniques, the researchers identified that thermodynamic lattice stability and geometric factors—not bulk stoichiometry—are the primary drivers of electronic properties in catalysts. This demonstrates how making machine learning models interpretable can reveal genuine scientific insights rather than just making accurate predictions.
-
Non-linear models substantially outperform traditional approaches: The machine learning framework achieved R² scores of 0.61-0.77 compared to R² of 0.32 for traditional linear methods, showing that complex structure-performance relationships in catalysis require more sophisticated modeling approaches to capture accurately.
-
Responsible AI can reduce experimental waste: By creating a framework that links catalyst design features to performance without expensive trial-and-error experimentation, this work demonstrates how responsible AI practices can accelerate discovery while reducing the resource consumption and environmental impact of the research process itself.
-
Transparency supports downstream integration: The researchers explicitly prioritized creating a "light-weight generalizable methodology" with ranked physical features that can be integrated into other models. This transparency and modularity makes the AI system more trustworthy and useful for domain experts who need to understand and build upon the results.
-
Limited data doesn't excuse poor interpretability: Despite working with constrained datasets, the authors chose interpretable tree-based models over black-box alternatives, showing that responsible AI practices—prioritizing explainability—should not be abandoned when data is scarce.
Abstract
The current catalyst discovery and development pipeline for energy-intensive applications like methane conversion remains bottlenecked by expensive trial-and-error experimentation, irreproducible chemical intuition, and a lack of frameworks linking complex catalytic design spaces to performance. This work presents an interpretable machine learning framework that integrates SHAP-based feature importance analysis (Explainable AI) with tree-based ensembles (Random Forest and Bayesian-optimized CatBoost) to characterize Fe-zeolite and oxide-supported catalysts for the partial oxidation of methane (POM). Despite limited data, the framework decodes complex structure-performance relationships by identifying and ranking thermodynamic, structural, and geometric microdescriptors that influence the electronic band gap and govern macroscale performance metrics such as selectivity, activity, and stability. This work explicitly demonstrates that thermodynamic lattice stability and geometric factors are the primary drivers of electronic band gap (a critical proxy for redox reactivity) rather than bulk stoichiometry. Non-linear models achieve an R2 of 0.61 - 0.77, significantly outperforming traditional linear baselines (R2 = 0.32). This workflow provides both a light-weight generalizable methodology and a prioritized list of physical features for accelerated catalyst screening - and these features can subsequently be integrated into microkinetic and reaction engineering models to create digital twins of complex reactor systems and to enable predictive optimization in autonomous R&D laboratories.
27 pages, 10 figures
Paper Info
arXiv ID2605.08994
PublishedMay 9, 2026
Citations
0
Relevance72%
VenuearXiv.org
TypeJournalArticle
Categoryphysics.chem-ph
Responsible AI Topics
ExplainabilityTransparencyAccountability
Links
Citation data from Semantic Scholar