r2VIM: A new variable selection method for random forests in genome-wide association studies

Table 2 Number of SNPs in simulation study 1 with empirical power > 0

Method	n	mtry	Factor	Total	Causal	High LD	Mod LD	Low LD	FP
LR	2000			40	8	15	11	4	0
LR	6000			98	9	16	15	15	0
r2VIM	2000	100000	1	38	7	13	10	2	4
r2VIM	2000	100000	3	28	7	12	6	2	0
r2VIM	2000	100000	5	24	7	10	5	2	0
r2VIM	2000	250000	1	40	8	12	9	2	8
r2VIM	2000	250000	3	25	7	9	5	2	1
r2VIM	2000	250000	5	23	7	9	5	2	0
r2VIM	6000	100000	1	51	9	16	10	5	3
r2VIM	6000	100000	3	41	9	16	6	4	0
r2VIM	6000	100000	5	37	9	16	6	4	0
r2VIM	6000	250000	1	63	9	16	12	6	13
r2VIM	6000	250000	3	42	9	16	7	4	1
r2VIM	6000	250000	5	37	9	16	5	4	0

Shown are results for logistic regression (LR) and r2VIM. Columns denote method, sample size (n), mtry parameter and factor for r2VIM, total number of SNPs, number of SNPs in strong (r ² > 0.8), moderate LD (0.5 < r ² ≤ 0.8) and low LD (0.3 < r ² ≤ 0.5) with any causal SNP as well as number of false-positive SNPs (FP)

ISSN: 1756-0381