Table 5. The accuracy of strategies for selecting events from NPRs depending on various parameters.

		as-median NPRs		random NPRs		s-median NPRs
Filtering Threshold T	Δ
0%	10%	11.3	20.3	11.3	20.1	10.1	17.7
	20%	11.8	20.0	11.5	20.3	11.5	16.0
50%	10%	10.0	18.0	9.8	17.4	8.2	14.9
	20%	9.3	15.9	8.9	15.7	7.5	10.9
90%	10%	8.5	15.2	8.4	15.1	7.7	13.5
	20%	7.1	11.4	7.0	11.4	7.1	9.4

Parameter T denotes the filtering threshold and Δ denotes the noise level for generating sets of NPRs. Columns 3, 5, and 7 report the accuracy as measured by the average symmetric distance to the event set in the true gene history. Note that the bold-faced values indicate the method (among (a)s-median and random NPRs) having the symmetric distances of 1000 gene families being significantly less than that of the other methods, i.e. p-value of the paired t.test being less than 0.05. Columns 4, 6 and 8 report the average numbers of predicted , , and events with or without filtering. On average, the true evolutionary history of a gene family contains 13.7 ,, and events.