In this review, we survey the fundamental problems and their solvers in molecular property prediction problems: graph neural networks modeling and uncertainty estimation. Molecules are naturally represented with graphs, thus graph neural networks, such as graph convolution networks and several types of message passing networks, are their natural models for accurate predictions. However, graph neural networks and any other deep learning models require a large amount of data. Unfortunately, molecular data are insufficient and skewed. Thus, not just an accurate method but models that return uncertainty estimations, such as Bayesian estimations, deep ensembles, and evidential methods, are required.
This work was supported by the Institute of Information & Communications Technology Planning Evaluation (IITP) grant funded by the Korean government (MSIT) (No.2022-0-00369) and by the National Research Foundation of Korea Grant funded by the Korean government (2022R1F1A1065664). *KH and JY contributed equally. \u2020LS is the corresponding author.