KNN-MDR: a learning approach for improving interactions mapping performances in genome wide association studies

BMC Bioinformatics

Table 2 Simulation results when G = 3 and the number of cases and controls is 500

Method	MDR	AntEpiSeeker	BOOST	MegaSNPHunter	KNN-MDR
Power	N/A	0.65	0.67	0.80	0.74
Corrected power	N/A	0.15	0.28	0.12	0.63

Tables 1 and 2 shows the results of 100 simulations. For KNN-MDR, the number of neighbors is set to 10, and the 1000 markers are split into 100 windows of 10 consecutive markers. All possible sets of up to 2 windows for Table 1 (5050 sets) and up to 3 windows for Table 2 (166750 sets) have been tested. Parameters for the other methods were set to default values. Due to the very large number of tests required when 3 markers are involved, MDR results have not been obtained in Table 2. Data sets used to generate these 2 tables are provided as Additional files 2 and 3

ISSN: 1471-2105