Skip to main content
. 2019 May 9;9:7155. doi: 10.1038/s41598-019-43664-y

Figure 4.

Figure 4

Performance (in terms of Average Precision) of the best BitterSweet models corresponding to each molecular descriptor set for (a) sweet/non-sweet prediction and (b) bitter/non-bitter prediction. Dragon2D/3D molecular descriptor set and Boruta feature selection were found to produce the most optimal models for sweet/non-sweet prediction. For bitter/non-bitter prediction, PCA outperformed Boruta.