Algorithm-Driven Robotic\nDiscovery of Polyoxometalate-Scaffolding\nMetal–Organic Frameworks

The experimental\nexploration of the chemical space of\ncrystalline\nmaterials, especially metal–organic frameworks (MOFs), requires\nmultiparameter control of a large set of reactions, which is unavoidably\ntime-consuming and labor-intensive when performed manually. To accelerate\nthe rate of material discovery while maintaining high reproducibility,\nwe developed a machine learning algorithm integrated with a robotic\nsynthesis platform for closed-loop exploration of the chemical space\nfor polyoxometalate-scaffolding metal–organic frameworks (POMOFs).\nThe eXtreme Gradient Boosting (XGBoost) model was optimized by using\nupdating data obtained from the uncertainty feedback experiments and\na multiclass classification extension based on the POMOF classification\nfrom their chemical constitution. The digital signatures for the robotic\nsynthesis of POMOFs were represented by the universal chemical description\nlanguage (χDL) to precisely record the synthetic steps and enhance\nthe reproducibility. Nine novel POMOFs including one with mixed ligands\nderived from individual ligands through the imidization reaction of\nPOM amine derivatives with various aldehydes have been discovered\nwith a good repeatability. In addition, chemical space maps were plotted\nbased on the XGBoost models whose F1 scores are above 0.8. Furthermore,\nthe electrochemical properties of the synthesized POMOFs indicate\nsuperior\nelectron transfer compared to the molecular POMs and the direct effect\nof the ratio of Zn, the type of ligands used, and the topology structures\nin POMOFs for modulating electron transfer abilities.