Table 2.

Influence of the replacement method on the Conserved Pairs Proportion (CPP). The numbers of clusters defined per metric are indicated (K^algo). The CPP and CPP_f values are given for the (1/7) subset of Ogawa [26] with missing values introduced (a) in the entire subsets and (b) only in extreme values (values > 1.5 or < -1.5). The CPP was computed by using the seven algorithms on the data without replacing the MVs (MV), after replacing the MVs by the kNN method (kNN) with the k_opt values (see Table 3), or, after replacing the MVs by zero (zero). For kNN and zero values, the CPP and CPP_f (with f = 5) values are given in (%). The percent of gain for all the simulations is indicated (+/-), i.e. the proportion of simulations with a better CPP value than the CPP value of MV. The values given in this table are the average values for MV in the range [1%; 50%] of genes with one MV per genes.

(a)
		CPP					CPPf
		MV	kNN		zero		MV	kNN		zero
Algorithm	K algo	(%)	(%)	(+/-)	(%)	(+/-)	(%)	(%)	(+/-)	(%)	(+/-)
single	175	96.4	98.3	99.3	98.2	99.6	98.8	99.1	53.7	99.0	47.8
centroid	175	95.3	96.0	74.8	95.7	64.1	97.7	97.9	57.6	97.8	53.3
average	83	73.2	75.3	66.3	74.8	62.0	92.0	92.5	54.4	93.0	57.6
median	165	70.1	71.5	55.2	72.2	54.4	85.2	86.0	52.8	86.2	52.1
mcquitty	48	58.2	59.1	54.5	58.6	50.1	74.9	75.3	51.9	75.1	49.5
Ward	19	51.0	52.4	62.4	50.7	49.1	71.8	73.3	60.2	71.5	49.1
complete	36	47.9	49.2	60.4	48.4	51.9	71.9	72.7	53.7	72.1	51.6
(b)
		CPP					CPPf
		MV	kNN		zero		MV	kNN		zero
Algorithm	K algo	(%)	(%)	(+/-)	(%)	(+/-)	(%)	(%)	(+/-)	(%)	(+/-)
single	175	97.9	98.4	98.5	98.7	98.3	98.8	99.3	86.9	99.7	76.6
centroid	175	96.5	96.7	64.8	96.8	60.6	98.5	98.5	55.9	97.7	32.8
average	83	81.4	81.0	45.1	83.3	66.1	93.9	94.5	57.0	94.9	61.3
median	165	75.0	74.2	45.7	74.2	47.9	87.0	86.5	50.3	85.1	45.4
mcquitty	48	71.6	71.3	48.1	71.9	51.0	83.1	83.2	49.6	84.6	58.2
Ward	19	63.8	62.4	42.1	60.8	29.9	83.0	81.1	37.1	79.8	27.0
complete	36	56.5	55.2	44.5	56.0	47.2	75.7	74.6	46.1	74.6	45.0