Adaptive symmetry-based adversarial perturbation augmentation for molecular graph representations with dual-fusion attention information

Shuting Jin, Xiangrong Liu, Junlin Xu, Sisi Yuan, Hongxing Xiang, Lian Shen, Chunyan Li, Zhangming Niu, Yinhui Jiang

Abstract: Highlights•An effective augmentation strategy can alleviate insufficient molecular labels.•Gapcl is a novel adversarial augmentation framework for molecular graph representations.•GapCL may help improve the generality of other molecular graph representation models.