Algorithm-Driven Robotic Discovery of Polyoxometalate-Scaffolding Metal-Organic Frameworks.

Area of Science:

• Materials Science
• Chemistry
• Artificial Intelligence

Background:

• Manual exploration of crystalline material chemical space, particularly metal-organic frameworks (MOFs), is time-consuming and labor-intensive.
• Accelerating material discovery requires efficient, reproducible synthesis and exploration methods.
• Polyoxometalate-scaffolding metal-organic frameworks (POMOFs) offer unique properties but their synthesis is complex.

Purpose of the Study:

• To develop a machine learning-integrated robotic platform for accelerated, closed-loop exploration of POMOF chemical space.
• To enhance the reproducibility and efficiency of novel POMOF synthesis.
• To investigate the electrochemical properties of newly discovered POMOFs and identify key structural modulators.

Main Methods:

• Integration of an eXtreme Gradient Boosting (XGBoost) machine learning model with a robotic synthesis platform for closed-loop material discovery.
• Optimization of the XGBoost model using uncertainty feedback and a multiclass classification extension for POMOFs.
• Utilized the universal chemical description language (χDL) for digital signatures to ensure precise recording and reproducibility of synthetic steps.

Main Results:

• Discovery of nine novel POMOFs, including one with mixed ligands, with high repeatability.
• Generation of chemical space maps based on XGBoost models with F1 scores above 0.8.
• Demonstrated superior electron transfer in synthesized POMOFs compared to molecular polyoxometalates (POMs).

Conclusions:

• The machine learning-driven robotic platform significantly accelerates the discovery of novel POMOFs.
• The ratio of Zn, ligand type, and topology structures are key factors in modulating POMOF electrochemical properties.
• This approach provides a robust framework for efficient and reproducible exploration of complex material chemical spaces.