. Author manuscript; available in PMC: 2019 Feb 13.

Published in final edited form as: Nat Genet. 2018 Aug 13;50(9):1335–1341. doi: 10.1038/s41588-018-0184-y

Table 1.

Comparison of different methods for GWAS with mixed effect models

		Method Features					Algorithm Complexity				Benchmarks for UK Biobank Data Coronary Artery Disease (PheCode 411)
		Does not require pre- computed genetic relationship matrix	Feasible for large sample sizes	Developed for binary traits	Accounts for unbalanced case- control ratio	Tests quantitative traits	Time complexity		Memory usage (Gbyte)		Time CPU hrs	Memory
		Does not require pre- computed genetic relationship matrix	Feasible for large sample sizes	Developed for binary traits	Accounts for unbalanced case- control ratio	Tests quantitative traits	Step 1	Step 2	Step 1	Step 2	Time CPU hrs	Memory
Logistic mixed model	SAIGE	✓	✓	✓	✓	✓	O(PM₁N^1.5) ^*	O(MN）	M₁N/4	N	517	10.3G
Logistic mixed model	GMMAT			✓		✓	O(PN³)	O(MN²)	F N²	F N²	NA	NA
Linear mixed model	BOLT-LMM	✓	✓			✓	O(PM₁N^1.5)^*	O(MN）	M₁N/4	N	360	10.9G
Linear mixed model	GEMMA					✓	O(N³)	O(MN²)	F N²	FN²	NA	NA

N: number of samples

P: number of iterations required to reach convergence

M₁: number of markers used to construct the kinship matrix;

M: total number of markers to be tested

F: Byte for floating number

Number of iterations in PCG is assumed as O(N^0.5)⁸