Figure 10

Inspecting the prediction error difference of two (Q)SAR approaches. The prediction error difference (error _{simple-linear} − _{support-vector} ) of two regression approaches for the Caco-2 dataset is selected. Simple linear regression performed especially bad for the two selected compounds (olsalazine and p n u200603). Both compounds have a very low logD value (logD is the top feature in the feature list on the right-hand side).