Skip to main content
Fig. 4 | Journal of Cheminformatics

Fig. 4

From: BitterMatch: recommendation systems for matching molecules with bitter taste receptors

Fig. 4

Performance of “new ligands” BitterMatch. A Average recall precision curves over 100 repetitions for the BitterMatch model, a prior model and a nearest neighbor model. 95% Bootstrap confidence intervals for the mean are shown for each model. B The number of tests required to find at least one receptor which is activated by the ligand. Each point represents a ligand in one of the test sets in 100 repetitions of the experiment. Experiments in which BitterMatch required less tests are shown in pink below the main diagonal, experiments in which the prior model required less tests are shown in yellow above the diagonal. Experiments in which the two methods required the same number of tests are shown in grey on the main diagonal. Darker points correspond to more experiments. The black dashed line is a fitted linear regression (for the number of tests by BitterMatch as a function of the number of tests by the prior) with intercept of 0.85 and coefficient of 0.28

Back to article page