- Keywords: Scientific Application, Knowledge Discovery, Graph Neural Network, Attention Mechanism
- TL;DR: We first propose a fully-automated and target-directed atomic importance estimator based on the graph neural networks and a new concept of reverse self-attention.
- Abstract: Estimating the importance of each atom in a molecule is one of the most appealing and challenging problems in chemistry, physics, and material engineering. The most common way to estimate the atomic importance is to compute the electronic structure using density-functional theory (DFT), and then to interpret it using domain knowledge of human experts. However, this conventional approach is impractical to the large molecular database because DFT calculation requires huge computation, specifically, O(n^4) time complexity w.r.t. the number of electrons in a molecule. Furthermore, the calculation results should be interpreted by the human experts to estimate the atomic importance in terms of the target molecular property. To tackle this problem, we first exploit machine learning-based approach for the atomic importance estimation. To this end, we propose reverse self-attention on graph neural networks and integrate it with graph-based molecular description. Our method provides an efficiently-automated and target-directed way to estimate the atomic importance without any domain knowledge on chemistry and physics.