Fig. 5.

Prediction rank for the molecules of the entire dataset (x-axis) vs out-of-sample predicted sweetness probability (y-axis). Reference molecule prediction are highlighted. SHAP profiles of two representative molecules: Sucrose (B) and Propanolol (C). For each figure, SHAP values are shown in the left panel and impacting feature distributions in the right panel, with values assumed by the features highlighted with solid red lines. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)