Skip to main content
Figure 4 | Journal of Cheminformatics

Figure 4

From: Proteochemometric modeling in a Bayesian framework

Figure 4

GP determine models applicability domain. The differences between the true and predicted bioactivities (y axis) and the errors on predictions estimated by the GP model (x axis) are compared for the adenosine receptor dataset with radial (A) and NP (B) kernel, and for the GPCRs dataset with radial (C) and NP (D) kernels. The distribution of the differences between true and predicted bioactivities increases with the GP error on the prediction. This validates that the GP error is a measurement of the Applicability Domain (AD) of the model.

Back to article page
\