A self-conformation-aware pre-training framework for molecular property prediction with substructure interpretability.

The major challenges in drug development stem from frequent structure-activity cliffs and unknown drug properties, which are expensive and time-consuming to estimate, contributing to a high rate of failures and substantial unavoidable costs in the clinical phases. Herein, we propose the self-conformation-aware graph transformer (SCAGE), an innovative deep learning architecture pretrained with approximately 5 million drug-like compounds for molecular property prediction. Notably, we develop a multitask pretraining framework, which incorporates four supervised and unsupervised tasks: molecular fingerprint prediction, functional group prediction using chemical prior information, 2D atomic distance prediction, and 3D bond angle prediction, covering aspects from molecular structures to functions. It enables learning comprehensive conformation-aware prior knowledge, thereby enhancing its generalization across various molecular property tasks. Moreover, we design a data-driven multiscale conformational learning strategy that effectively guides the model in understanding and representing atomic relationships at the molecular conformational scale. SCAGE achieves significant performance improvements across 9 molecular properties and 30 structure-activity cliff benchmarks. Case studies demonstrate that SCAGE accurately captures crucial functional groups at the atomic level, which are closely associated with molecular activity, providing valuable insights into quantitative structure-activity relationships.