Figure 2

Noise influence in model performance. RMSEP_ext (red) and $R_0^2\mathit{\text{ext}}$ (black) values obtained when increasing the noise level (noise variance added to the diagonal of the covariance matrix) were calculated for: adenosine receptors (left figure), GPCRs (medium figure) and dengue virus NS3 proteases (right figure). Upper plots correspond to GP models calculated with the radial kernel while the bottom plots refer to GP models with the normalized polynomial (NP) kernel. In all cases, the radial kernel appears more sensitive to noise, while the NP kernel performs equally well when noise is added to the data. These data suggest that the NP kernel is more appropriate for the modeling of noisy PCM datasets.