TransPolymer: a Transformer-based language model for polymer property predictions

Accurate and efficient prediction of polymer properties is crucial for polymer design. This paper introduces TransPolymer, a Transformer-based language model for polymer property prediction. The model uses a chemically aware tokenizer to learn representations from polymer sequences, showing superior performance on ten polymer property prediction benchmarks. Pretraining on a large unlabeled dataset via Masked Language Modeling further enhances its performance, highlighting the importance of self-attention in modeling polymer sequences. TransPolymer is presented as a promising computational tool for rational polymer design and understanding structure-property relationships.

Changwen Xu, Yuyang Wang, Amir Barati Farimani