Graph-pKₐ is a quantitative structure-property relationship
(QSPR) model to predict acid dissociation constant (pKₐ) of small
molecules in aqueous solution. Combining multi-instance learning into
graph neural network, Graph-pKₐ not only performs well in
predicting macro pKₐ, but more significantly, can learn the micro
pKₐ values of atoms through training against the macro pKₐ
values of molecules.
Here, we provided a Web version of Graph-pKₐ model that was trained with about 17000 pKₐ datapoints.