A new protein binding pocket similarity measure based on comparison of clouds of atoms in 3D: application to ligand prediction

Table 4 Classification error on the extended Kahraman benchmark

Method	A	B	C
sup-CK_L	0.19	0.21	0.18
sup-CK_L-Vol	0.18	0.19	0.18
Vol	0.32	0.39	0.31
Princ-Axis	0.22	0.27	0.21
sup-PI	0.24	0.33	0.23

Classification error for all algorithms on the extended Kahraman dataset. Column A - classification error evaluated on all 972 pockets. Column B Proportion of wrong predictions among the original 100 Kahraman pockets extracted from column A, i.e. classification error evaluated on 100 Kahraman pockets when all 972 pockets are used in the leave-one-out procedure. Column C - classification error evaluated on the 872 new pockets, when the 100 Kahraman pockets are not used in the leave-one-out procedure.

ISSN: 1471-2105