Feature attributions for water-solubility predictions obtained by artificial intelligence methods and chemists

Recently, the field of explainable artificial intelligence has attracted significant research interest, with a particular focus on “feature attribution” in the field of chemistry. However, studies comparing the relationship between artificial-intelligence- and human-based feature attributions when predicting the same outcome are scarce. Hence, the current study aims to investigate this relationship by comparing machine-learning-based feature attributions (graph neural networks and integrated gradients) with those of chemists (Hansch–Fujita method) when predicting water solubility. The findings reveal that the artificial-intelligence-based attributions are similar to those of chemists despite their distinct origins.