I. Introduction
Molecule property prediction is one of the fundamental tasks in chemistry [1]. Traditional computational methods, such as density functional theory (DFT), are time-consuming and poorly scalable with size [2]. Recently, machine learning (ML), including deep learning (DL), has emerged as a powerful data-driven approach for establishing the correlation between molecular structure and properties. ML methods can sometimes deliver results with precision comparable with DFT but be 3–5 orders of magnitude faster [2]–[4].