Our work on multi-stage machine learning for catalyst discovery is now published in Science Advances.

The paper presents an integrated framework combining data mining, active learning, and domain adaptation for efficient screening of acidic OER electrocatalysts. Starting from literature-mined data, the pipeline progressively narrows the candidate space through ML-guided selection, ultimately identifying experimentally validated high-performance materials.

Key results:

• Systematic navigation of multimetallic catalyst design space
• Experimental validation of ML-predicted candidates
• Demonstration of cross-domain transfer from alkaline to acidic conditions

Science Advances 2025, 11, eadr9038