Periodicity-aware deep learning for polymers

Deep learning has revolutionized chemical research by accelerating the discovery of new substances and enhancing the understanding of complex chemical systems. However, polymer chemistry, one of the most active branches of chemistry, has yet to establish a unified deep learning framework due to the complexity of polymer structures. Existing self-supervised learning methods for polymers simplify them into repeating units and neglect their inherent periodicity, thereby limiting the models’ ability to generalize across various tasks. To address this challenge, we herein propose a periodicity-aware deep learning framework for polymers (named PerioGT). In pre-training, a chemical knowledge-driven periodicity prior is constructed and incorporated into the model through contrastive learning. Then, periodicity prompts are learned in fine-tuning based on the periodicity prior to better leveraging the knowledge acquired in pre-training. Additionally, a novel graph augmentation strategy is employed for polymers, which integrates additional conditions via virtual nodes to effectively model complex chemical interactions. PerioGT achieves state-of-the-art performance on 12 downstream tasks. Moreover, wet-lab experiments in antimicrobial polymer discovery highlight PerioGT’s potential in the real world, identifying two polymers with potent antimicrobial properties. All the results demonstrate that introducing the periodicity prior of polymers effectively improves the model performance.