Figure 3

Analysis of the confidence intervals predicted on (left) the adenosine receptors and (right) aminergic GPCRs external sets. The percentages of annotated values lying within the intervals of confidence of 68%, 80%, 95% and 99% (ordinate axis) are depicted versus the size of the intervals. The blue line defines the theoretical proportionality between the size of confidence intervals and the number of points within the intervals, in the frame of the Gaussian cumulative function. The radial, PUK, and normalized polynomial (NP) kernels are in close conformity with the cumulative Gaussian distribution in both datasets, while the Laplacian and Bessel exhibit a diverse behavior depending on the dataset. Therefore, GP provide prediction errors in agreement with the Cumulative Gaussian distribution which can be reliably used to define intervals of confidence for the predictions.