Table 1:

Simulation study datasets. 30 replicates of each configuration were generated. Model architecture difficulties are designated by ‘E’ (easy), and ‘H’ (hard). Simulation method generation is designated as either ‘G’ (GAMETES), ‘C’ (custom script), or ‘G+C’ (GAMETES modified by custom script).

Simulated Data Group Description or Pattern of Association	Configurations	Config. Variations	Predictive Features	Total Features	Model Difficulty	Heritability	Instances	Simulation Method
2-way Pure Epistais (Core Datasets) Others marked by ‘*’	32	-	2	20	E, H	0.05, 0.1, 0.2, 0.4	200, 400, 800, 1600	G
1-Feature Main Effect	8	-	1	20	E,	0.05,	1600	G
					H	0.1,
						0.2,
						0.4
2-Feature Additive Effect	2	50:50, 75:25	2	20	E	0.4	1600	G
4-Feature Additive Effect	1	-	1	20	E	0.4	1600	G
4-Feat. Additive	2	50:50,	2	20	E	0.4	1600	G
2-way Epistasis		75:25
4-Feat. Heterogeneous	2	50:50,	2	20	E	0.4	1600	G
2-way Epistasis		75:25
3-way Pure Epistasis	1	-	3	20	E	0.2	1600	G
Number of Features*	4		2	100, 1000, 10000, 100000	E	0.4	1600	G
Continuous Features*	1	-	2	20	E	0.4	1600	G+C
Mix of Discrete and Continuous Features*	1		2	20	E	0.4	1600	G+C
Continuous Endpoint*	3	0.2, 0.5, 0.8	2	20	E	0.4	1600	G
Continuous Endpoint* (1-Threshold Model)	1	“	2	20	E	0.4	1600	G+C
Missing Data*	4	0.001, 0.01,	2	20	E	0.4	1600	G+C
		0.1, 0.5
Imbalanced Data*	2	0.6, 0.9	2	20	E	0.4	1600	G
Multi-class Endpoint	2	3-class,	2	20	N/A	1	1600	C
(Impure 2-way Epistasis)		9-class
XOR Model	4	2-way,	2	20	N/A	1	1600	C
(Pure Epistasis)		3-way,	3
		4-way,	4
		5-way	5
Multiplexer (MUX)	6	6-bit →	2	6	3-way	1	500	C
(Pure Epistasis and		11-bit →	3	11	4-way		1000
Heterogeneous		20-bit →	4	20	5-way		2000
Associations)		37-bit →	5	37	6-way		5000
		70-bit →	6	70	7-way		10000
		135-bit →	7	135	8-way		20000