An integrated approach with new strategies for QSAR models and lead optimization

BMC Genomics

Table 2 Accuracies of our method using different protein-ligand interaction profiles on huAChE set

	All interaction profile^a		Consensus feature profile^b
	GEMPLS	GEMkNN	GEMPLS	GEMkNN	Specific^g Skeleton
No. of total features (atoms)	223	223	78	156	92
No. of selected features (atoms)	36.2	36.1	22	29.8	23.5
Average of q ² (Training)^c	0.627	0.657	0.658	0.704	0.817
Average of r ² (Testing)^d	0.402	0.123	0.467	0.063	0.723
Standard derivation of q ^2e	0.015	0.018	0.004	0.009	0.006
Standard derivation of r ^2f	0.125	0.095	0.085	0.050	0.056

^a,bUsing all and only consensus interaction profiles as descriptions, respectively
^c,dThe average q ² and r ² values of 30 times for the training set and testing set, respectively
^e,fThe standard deviation q ² and r ² values of 30 times for the training set and testing set, respectively
^gUsing the interaction profiles of both consensus feature profile and specific skeleton

ISSN: 1471-2164