A method combining a random forest-based technique with the modeling of linkage disequilibrium through latent variables, to run multilocus genome-wide association studies

Table 6 Range of variation, average and standard deviation for the distributions of variable importances across the SNPs, obtained from T-Trees and the hybrid FLTM / T-Trees approach. Excerpt of Table 1 in Additional file 1 (Supplementary data)

Pathology	Method	Minimum	Maximum	Average	Standard
Chromosome					deviation
TD2, Chr10	T-Trees	0	0.019	3.8e-05	2.5e-04
	hybrid	0	0.089	3.2e-05	6.4e-04
RA, Chr19	T-Trees	0	0.372	1.7e-04	4.9e-03
	hybrid	0	0.391	2.1e-04	5.3e-03
CD, Chr20	T-Trees	0	0.030	5.9e-05	8.6e-05
	hybrid	0	0.119	8.6e-05	1.2e-03

Convention : 0 stands for non-detectable at 10⁻⁸ threshold. The two first datasets are representative of the general trend observed over the 14 datasets analyzed: same order of magnitude for maxima, averages and standard deviations, respectively. Besides, dataset (T2D, Chr10) is among the datasets showing the smallest maxima. Dataset (RA, Chr19) shows the highest maxima. (CD, Chr20) represents the unique case of discrepencies between the orders of magnitude of the maxima (in favor of the hybrid approach). Dataset (CD, Chr20) is also one of the three cases showing a discrepancy in the standard deviations’ orders of magnitude

ISSN: 1471-2105