Table 5. Comparison of RFs induced using non-redundant subsets of the Kinase dataset.

Threshold	Non-redundant Observations (Pos/Unl)	Non-redundant Dataset G Mean	Entire Dataset
			TP	FP	TN	FN	G Mean
20%	102 (18/84)	0.79	51	196	371	43	0.60
30%	198 (26/172)	0.85	49	165	402	45	0.61
40%	332 (49/283)	0.78	75	184	383	19	0.73
50%	432 (67/365)	0.79	72	120	447	22	0.78
60%	497 (77/420)	0.81	77	132	435	17	0.79
70%	569 (83/486)	0.79	72	118	449	22	0.78
80%	625 (88/537)	0.80	72	112	455	22	0.78
90%	650 (94/556)	0.79	69	90	477	25	0.79
100%	661 (94/567)	0.80	72	98	469	22	0.80

For each threshold, a non-redundant dataset was generated using Leaf and used to induce a RF. The RF was then used to classify the proteins in both the non-redundant dataset it was trained on and the entire Kinase dataset. The TPs/FNs are the number of positive proteins in the entire dataset predicted correctly/incorrectly, and the TNs/FPs are the number of unlabelled proteins predicted correctly/incorrectly.