Chemically Transferable\nElectronic Coarse Graining\nfor Polythiophenes

Recent advances in machine-learning-based electronic\ncoarse graining\n(ECG) methods have demonstrated the potential to enable electronic\npredictions in soft materials at mesoscopic length scales. However,\nprevious ECG models have yet to confront the issue of chemical transferability.\nIn this study, we develop chemically transferable ECG models for polythiophenes\nusing graph neural networks. Our models are trained on a data set\nthat samples over the conformational space of random polythiophene\nsequences generated with 15 different monomer chemistries and three\ndifferent degrees of polymerization. We systematically explore the\nimpact of coarse-grained representation on ECG accuracy, highlighting\nthe significance of preserving the C-β coordinates in thiophene.\nWe also find that integrating unique polymer sequences into training\nenhances the model performance more efficiently than augmenting conformational\nsampling for sequences already in the training data set. Moreover,\nour ECG models, developed initially for one property and one level\nof quantum chemical theory, can be efficiently transferred to related\nproperties and higher levels of theory with minimal additional data.\nThe chemically transferable ECG model introduced in this work will\nserve as a foundation model for new classes of chemically transferable\nECG predictions across chemical space.