UNCLES: method for the identification of genes differentially consistently co-expressed in a specific subset of datasets

Table 1 Clustering methods’ performance comparison

C	GS	UNCLES versus closest competitor^*	Most separated pair of other methods^*
C1	1,200	−0.81 ± 0.15 (9.3 × 10⁻⁶¹) [HC]	−0.13 ± 0.17 (1.5 × 10⁻¹⁰) [HC, RV]
	2,000	−0.88 ± 0.17 (7.3 × 10⁻⁵⁵) [HC]	−0.15 ± 0.18 (1.7 × 10⁻¹¹) [HC, RV]
	3,000	−0.93 ± 0.15 (1.6 × 10⁻⁶⁸) [HC]	−0.12 ± 0.16 (2.5 × 10⁻¹¹) [SOMs, RV]
	5,000	−0.92 ± 0.15 (7.6 × 10⁻⁶⁶) [HC]	−0.09 ± 0.14 (1.9 × 10⁻⁸) [SOMs, RV]
	7,000	−0.77 ± 0.15 (3.6 × 10⁻⁵⁴) [SOMs]	−0.08 ± 0.12 (2.9 × 10⁻⁹) [SOMs, RV]
C2	1,200	−0.93 ± 0.15 (<10⁻²⁵⁵) [SOMs]	−0.04 ± 0.14 (5.8 × 10⁻⁷) [SOMs, RV]
	2,000	−0.92 ± 0.17 (<10⁻²⁵⁵) [HC]	−0.04 ± 0.12 (5.0 × 10⁻⁷) [HC, RV]
	3,000	−0.60 ± 0.15 (6.3 × 10⁻²⁴⁴) [HC]	−0.03 ± 0.11 (6.7 × 10⁻⁵) [HC, RV]
	5,000	−0.55 ± 0.13 (1.1 × 10⁻²³⁴) [HC]	−0.02 ± 0.09 (2.0 × 10⁻⁴) [HC, RV]
	7,000	−0.48 ± 0.13 (4.8 × 10⁻²¹⁹) [HC]	−0.02 ± 0.09 (1.3 × 10⁻³) [HC, RV]

*The format of the entries in these two columns is: μ ± σ (p-value) [method(s)]. The closest competitor to UNCLES is the one with the largest p-value while the most significantly separated pair of other clustering methods is the pair with the smallest p-value. See Methods for details

ISSN: 1471-2105