Representing Polymers as Periodic Graphs with Learned Descriptors for Accurate Polymer Property Predictions

Abstract

Accurately predicting new polymers' properties with machine learning models apriori to synthesis has potential to significantly accelerate new polymers' discovery and development. However, accurately and efficiently capturing polymers' complex, periodic structures in machine learning models remains a grand challenge for the polymer cheminformatics community. Specifically, there has yet to be an ideal solution for the problems of how to capture the periodicity of polymers, as well as how to optimally develop polymer descriptors without requiring human-based feature design. In this work, we tackle these problems by utilizing a periodic polymer graph representation that accounts for polymers' periodicity and coupling it with a message-passing neural network that leverages the power of graph deep learning to automatically learn chemically relevant polymer descriptors. Remarkably, this approach achieves state-of-the-art performance on 8 out of 10 distinct polymer property prediction tasks. These results highlight the advancement in predictive capability that is possible through learning descriptors that are specifically optimized for capturing the unique chemical structure of polymers.

Polymer Informatics at Scale with Multitask Graph Neural Networks

“…A handful of polymer GNNs have been explored in the past. − The majority of these approaches are single task. The GNN proposed by Mohapatra et al is suitable for biopolymers, in which the monomer sequence is known.…”

“…The GNN proposed by Mohapatra et al is suitable for biopolymers, in which the monomer sequence is known. Other approaches, ,,, geared toward synthetic polymers (the subject of interest in this work), represent a polymer using the graph of a predominant repeat unit. This introduces the need for invariance to certain transformations of the repeat unit graph: translation, addition, and subtraction (as defined in Section ).…”

“…This introduces the need for invariance to certain transformations of the repeat unit graph: translation, addition, and subtraction (as defined in Section ). A subset of the GNNs for synthetic polymers , are invariant to translation, but not to addition and subtraction. In other words, a GNN that preserves the invariant properties of polymer repeat units has not been developed until now.…”

Augmenting Polymer Datasets by Iterative Rearrangement

“…One of the most promising solutions that capture the repetitive or periodic nature of a polymer is the periodic graph representation where the two terminal ends of the polymer chain are connected, forming a ring. , Although this representation is not physically accurate, the periodic polymer graph effectively captures the bonding environment and periodicity of a polymer, outperforming the monomer, dimer, and trimer graph in several unique polymer prediction tasks …”

“…Data augmentation has been extremely successful in computer vision and natural language processing , because it maximizes the amount of information that can be learned from scarce data by performing transformations which allow ML models to be insensitive to translation, rotation, and permutation variances. For example, there are permutation variances in the representation of a polymer due to translations along the polymer backbone. , Models trained on augmented data can be less sensitive to transformations in real (non-augmented) data . Namely, SMILES representations can be augmented with new noncanonical SMILES from the same molecule by changing the order in which the molecular graph is traversed each time. , Augmenting SMILES has been shown to improve the property prediction performance of a long short-term memory (LSTM) neural network (NN) because the model can learn the grammar of SMILES notation…”