Multi-instance learning of graph neural networks for aqueous pK(a) prediction

MOTIVATION: The acid dissociation constant (pK(a)) is a critical parameter to reflect the ionization ability of chemical compounds and is widely applied in a variety of industries. However, the experimental determination of pK(a) is intricate and time-consuming, especially for the exact determination of micro-pK(a) information at the atomic level. Hence, a fast and accurate prediction of pK(a) values of chemical compounds is of broad interest. RESULTS: Here, we compiled a large-scale pK(a) dataset containing 16 595 compounds with 17 489 pK(a) values. Based on this dataset, a novel pK(a) prediction model, named Graph-pK(a), was established using graph neural networks. Graph-pK(a) performed well on the prediction of macro-pK(a) values, with a mean absolute error around 0.55 and a coefficient of determination around 0.92 on the test dataset. Furthermore, combining multi-instance learning, Graph-pK(a) was also able to automatically deconvolute the predicted macro-pK(a) into discrete micro-pK(a) values. AVAILABILITY AND IMPLEMENTATION: The Graph-pK(a) model is now freely accessible via a web-based interface (https://pka.simm.ac.cn/). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.