. Author manuscript; available in PMC: 2011 Oct 25.

Published in final edited form as: J Chem Inf Model. 2010 Oct 25;50(10):1759–1771. doi: 10.1021/ci100200u

Table 5.

Performance of logistic regression models by recursive elimination of descriptors.

Top Descriptors	Full Model	LOO-CV	Delta
12	69/70 (0.98571)	33/70 (0.47143)	0.51429
11	64/70 (0.91429)	33/70 (0.47143)	0.44286
10	60/70 (0.85714)	34/70 (0.48571)	0.37143
9	55/70 (0.78571)	32/70 (0.45714)	0.32857
8	54/70 (0.77143)	31/70 (0.44286)	0.32857
7	51/70 (0.72957)	35/70 (0.5)	0.22857
6	52/70 (0.74296)	38/70 (0.54286)	0.2
5^a	50/70 (0.71429)	41/70 (0.58571)	0.12857
4	49/70 (0.7)	40/70 (0.57143)	0.12857
3	46/70 (0.65714)	37/70 (0.52857)	0.12857
2	37/70 (0.52857)	32/70 (0.45714)	0.071429

The model utilizing the top five descriptors is considered to be the optimal model with its low overfitting effects and good predictability based on leave-one-out cross validation.