A Phased Multimodal Machine Learning Framework for Accelerated Prediction of Cyclic Olefin Copolymerization Performance from Catalyst Complexes

A phased multimodal machine learning framework is established for the accelerated prediction of cyclic olefin copolymerization performance by half-metallocene catalysts. Data augmentation is performed on the sample set based on a kinetic model, which is developed by the method of moments. Feature expansion integrates interactive quantum chemical descriptors, quantum chemical descriptors, and reaction conditions to comprehensively capture quantitative structure–property relationships. Phase I model maps ligand topologies to complex properties (R 2 = 0.9592), expanding the ligand library with ethylene homopolymerization data to construct a virtual chemical space of 1,554 catalysts; Phase II model correlates these properties with reaction conditions to predict catalytic activity, number-average molecular weight, and comonomer content (R 2 = 0.9937). The framework achieves rapid prediction from catalyst libraries and reaction conditions to polymer properties. This methodology is applicable to other systems, including metallocenes, showing broad application prospects in polyolefin catalysis.