While feature importance analysis in chemical and materials machine learning can often be sensitive to both the predictive model and the attribution rule, the robustness of these rankings is rarely quantified before they are used to gain physical insights. Here, we compare 26 feature importance pipelines spanning data-driven, model-based, and formula-based analyses on a metal-support interaction data set anchored by an explicit SISSO equation, and we examine whether the same qualitative behavior recurs in high-entropy-alloy and halide perovskite data sets. Across the three benchmarks, we observe high intrafamily agreement but substantial interfamily variance. While a small subset of features remains stable across multiple families, several midranked features are highly family dependent, with their apparent importance shifting according to the underlying modeling assumptions. To ensure robust interpretability, we recommend that feature importance be reported by method family or correlation-based clusters, supplemented by resampling intervals.
Ruilin Lai, Xiaotong Liu, Yuhang Wang et al.· Journal of Chemical Informat...· 0 citations
The active sites of alloy catalysts may emerge only under reaction conditions, yet how reactive atmospheres select these sites remains unresolved. Here, we reveal how reactive atmospheres reorganize complex alloy surfaces into functional active-state distributions by using a machine-learning-potential-accelerated multiscale framework. Using selective acetylene hydrogenation on Pd–Ag alloys as a model reaction, we show that adsorbates reverse the intrinsic Ag-segregation tendency, enrich Pd in the outermost layer, and generate a distribution of Pd3 hollow ensembles with distinct second-shell coordination environments. These dynamically formed ensembles, rather than the as-prepared isolated Pd sites, govern the calculated activity and selectivity trends. Ensemble-resolved energetics and kinetics identify the adsorption free-energy gap between ethylene and acetylene as a predictive descriptor that captures the activity-selectivity trade-off and defines an optimal window balancing acetylene hydrogenation and ethylene desorption. Extending the analysis across multiple alloy families further reveals a general scaling relationship in which adsorbate-driven self-organization is governed by adsorption asymmetry and alloy stability. These results establish dynamic ensemble selection as a transferable framework for understanding and designing adaptive alloy catalysts, shifting catalyst optimization from static structural descriptors toward reaction-condition-directed control of active-state distributions.
Hong-Yue Wang, Hao Wang, Wei-Xue Li et al.· JACS Au· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
