Accelerating Multi-Elemental Catalyst Discovery with Interpretable Machine Learning and Automated Experimentation.

Area of Science:

• Catalysis
• Materials Science
• Environmental Chemistry

Background:

• Multielemental catalysts (MECs) offer tunable properties but are challenging to optimize using traditional methods.
• Trial-and-error synthesis and testing limit practical advancements in MEC development.

Purpose of the Study:

• To accelerate the development of MECs for tetracycline degradation using a peroxymonosulfate (PMS)-based Fenton-like system.
• To implement a reproducible, closed-loop discovery workflow integrating robotics and machine learning.

Main Methods:

• A robotic automation stack was coupled with a machine-learning-guided optimization loop.
• An adaptive-learning genetic algorithm (GA) trained a multilayer perceptron (MLP) surrogate model.
• Inductively coupled plasma optical emission spectrometry and SHapley Additive exPlanations (SHAP) were used for analysis.

Main Results:

• The GA-MLP closed loop identified four high-performing MECs, increasing tetracycline degradation efficiency from ~78% to 93%.
• Robotic and manual synthesis protocols yielded consistent catalyst performance.
• SHAP analysis provided quantitative insights into composition-performance relationships.

Conclusions:

• The implemented workflow is reproducible, automation-ready, and sample-efficient for optimizing MECs.
• This strategy is effective for advanced oxidation processes in environmental applications.
• The study demonstrates a viable digital approach for catalyst discovery and optimization.