Figure 1.
Flowchart of the QSAR study.
Figure 2.
Influence of the number of descriptors on the square of Pearson correlation coefficient (R2), leave one out (LOO) cross-validation coefficient (Rcv2), and F-values of the BMLR models.
Figure 3.
Crystal structure of EGFR bound to the 4-anilinoquinazoline inhibitor Erlotinib (PDB ID: 1M17).
EGFR is represented by a cartoon model, with the side chains of the binding site wire-framed in cyan and labeled. The inhibitor is represented by a stick-and-ball model, where carbons are colored in yellow, nitrogens in blue, oxygens in red, and hydrogens in light grey. The hydrogen bond between an amide nitrogen donor of the ligand and the carboxyl group of MET769 in the receptor is plotted as a red dotted line. Figure produced with the PyMOL program [56].
Figure 4.
Heat maps of the correlation matrix of the molecular descriptors.
Rows and columns are ordered according to a hierarchical clustering (cluster tree lines on the side and top) of the selected molecular descriptors codes.
Figure 5.
Plot for the training and test EGFR inhibitors of the pIC50 values predicted by the BMLR method versus the experimental pIC50 values.
Table 1.
Comparison of the statistical results between the BMLR and GS-PPR models.
Figure 6.
Principal component analysis of the nine selected descriptors.
Arrows indicate the directions of the variable vectors in the principal component space. Black circles denote compounds from the training set, blue circles those from the test set.
Figure 7.
Statistical parameters of the training set and the test set during the optimization process of the PPR parameters by the grid search method.
(a): R2, (b): RMSE.
Figure 8.
Plot for the training and test EGFR inhibitors of the pIC50 values predicted by the GS–PPR method versus the experimental pIC50 values.