Answers are grounded in this paper and cite their sources.
Structure–activity relationships built on structure-derived descriptors are a dominant paradigm in catalyst screening. However, this structure-driven framework scales poorly to multimetallic systems, where active sites are heterogeneous, dynamically reconstruct under operating conditions, and require extensive structural enumeration and system-specific descriptor engineering. Here, we introduce a composition-driven screening strategy for multimetallic catalysts using feature construction inspired by the virtual crystal approximation. Specifically, multimetallic sites are represented as composition-weighted pseudoatoms, allowing readily available elemental properties to be converted into predictive features without explicit structural modeling. Unlike elemental fractions alone, these features encode physicochemical differences among the constituent metals. We demonstrate this approach using Prussian blue analogues (PBAs) as a compositionally tunable and reconstruction-prone validation platform for the oxygen evolution reaction (OER). Combining the composition-derived features with symbolic regression yields an explicit expression for ranking compositions, guiding experimental identification of four high-performance PBAs. We further evaluate this strategy across six literature data sets spanning different catalyst families and reactions, independently training a model for each. Using one of the models, we screen and experimentally validate a perovskite catalyst achieving an overpotential of 189 mV at 10 mA cm–2 in acidic OER. These results demonstrate the broad applicability of virtual crystal approximation-inspired feature construction for composition-driven screening of multimetallic catalysts.
No takes yet. Share an insight, caveat, or question.
Zeng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: