Table 3.

Exploration of descriptor selection using the RF algorithm with ExtCDK FP and 1D&2D descriptors for the training set.

Model	SE 1	SP_B 2	SP_C 3	Q 4	MCC 5
ExtCDK FP
50 6	0.41	0.19	0.65	0.59	0.21
100 6	0.44	0.14	0.70	0.64	0.23
150 6	0.46	0.17	0.71	0.64	0.27
200 6	0.44	0.14	0.71	0.65	0.23
1D&2D descriptors
50 6	0.54	0.11	0.69	0.63	0.25
100 6	0.59	0.09	0.70	0.64	0.25
150 6	0.55	0.12	0.69	0.63	0.26
200 6	0.55	0.08	0.71	0.65	0.22

¹ Sensitivity, the ratio of true A to the sum of true A and false A. ² Specificity B, the ratio of true B to the sum of true B and false B. ³ Specificity C, the ratio of true C to the sum of true C and false C. ⁴ Overall predictive accuracy, the ratio of the sum of true A, true B, and true C to the sum of true A, true B, true C, false A, false B, and false C. ⁵ MCC, Matthews correlation coefficient. ⁶ The most important descriptors in accordance with the mean decrease in accuracy.