Molecular Topological Deep Learning for Polymer Property Prediction

Accurate and efficient prediction of polymer properties is of key importance for polymer design. Traditional experimental tools and density functional theory (DFT)-based simulations for polymer property evaluation are both expensive and time-consuming. Recently, a gigantic amount of graph-based molecular deep learning models has emerged and demonstrated huge potential in molecular data analysis. Even with great progress, these models tend to ignore the high-order and multiscale information within the data. In this paper, we developed molecular topological deep learning (Mol-TDL) for polymer property analysis. Our Mol-TDL incorporates both high-order interactions and multiscale properties into a topological deep learning architecture. The key idea is to represent polymer molecules as a series of simplicial complexes at different scales and build up simplicial neural networks accordingly. The aggregated information from different scales provides a more accurate prediction of the polymer molecular properties. Further, we develop a multiscale topological contrastive learning model and use it for pretraining of simplex-based message passing. Our model has been extensively tested on both well-established DFT-based and experimental polymer data sets with various polymer properties. It has been found that our model can outperform all existing learning models, as far as we know. More importantly, our Mol-TDL model has been used in the prediction of glass transition temperatures for eight advanced polymers. We synthesized these polymers and measured their glass transition temperatures. Our model can achieve highly accurate predictions with a mean error of about 45 °C, demonstrating the huge potential of our Mol-TDL model in polymer design and discovery.