Machine-learning methods are gaining traction in the design of aqueous zinc-ion battery (AZIB) cathodes, yet they remain hampered by the inefficiency and high cost of trial-and-error experimental screening. Pinpointing the key parameters of metal-doped MnO₂ and predicting their electrochemical performance in a single step has long been a challenge in the field.
This study breaks free from traditional experimental loops by constructing, for the first time, a literature-plus-experiment data set covering 21 metal dopants. Elemental descriptors, synthesis routes and testing conditions are coupled into a machine-learning pipeline. After feature-engineering-based dimensionality reduction, an extreme gradient boosting (XGB) model achieves a predictive R² of 0.92 on the test set. SHAP interpretability analysis further identifies “current density–dopant ratio–molecular weight” as the three decisive factors governing specific capacity.
Guided by the model, the team synthesized Fe- and Ni-doped MnO₂ and measured rate performance. The Ni-doped sample retains a better rate performance than the Fe-doped one. Furthermore, DFT calculations show the Ni-doped structure’s band gap narrows by 0.08 eV relative to the Fe-doped structure, in close agreement with model predictions.
The researchers have also released an online prediction platform that returns a specific-capacity estimate within seconds after users input elemental, synthesis and testing parameters. The framework offers an expandable, interactive paradigm for “compute-first, experiment-later” rational design of AZIB cathodes. It can be extended to high-throughput screening of other metal-doped energy-storage materials. The work titled “A data-driven approach for rapid revealing of metal doping in MnO₂ cathodes for high-performance aqueous zinc-ion batteries”, was published on Acta Physico-Chimica Sinica (published on Dec. 24th, 2025).

DOI: 10.1016/j.actphy.2025.100232