Figure 4.

(a) Model for predicting basis set superposition error trained with 312 protein backbone-backbone hydrogen bonding fragment interactions. The data were fit with a linear regression model using a bimolecular proximity descriptor as an independent variable and had an R² of 0.89. (b) The model for BSSE trained with 354 nonpolar complexes from the protein fragment database yielded an R² of 0.85.