Fig. 1

Schematic workflow of a two-dimensional prediction of the odor of a molecule using same-weighted OWSum. A training set contains molecules together with their descriptors (here floral and medicinal) and extracted features that are structural patterns. For simplicity, we only regard three features. Based on the training set, OWSum calculates the influence I by multiplying the weight G with the weighting factor a (here 1 as we use the same-weighted OWSum). For the prediction of a molecule, all features that occur in that molecule are considered, in this case the first and the second feature ([CX4H3] and [CX4]). By summing up their influence, OWSum calculates one score per descriptor. As the score for floral (1.67) is higher than the score for medicinal (1.50), OWSum predicts the odor floral. As floral is in fact the odor of the molecule, the prediction is accepted as correct. See the Methods section for a detailed explanation of the algorithm. (Created with BioRender.com)