Interpretable Attribution Assignment for Octanol-Water Partition Coefficient

With the increasing development of machine learning models,theircredibility has become an important issue. In chemistry, attributionassignment is gaining relevance when it comes to designing moleculesand debugging models. However, attention has only been paid to whichatoms are important in the prediction and not to whether the attributionis reasonable. In this study, we developed a graph neural networkmodel, a highly interpretable attribution model in chemistry, andmodified the integrated gradients method. The credibility of our approachwas confirmed by predicting the octanol-water partition coefficient(logP) and evaluating the three metrics (accuracy, consistency, andstability) in the attribution assignment.