Table 3. A comparison of Dsuite and a number of other tools in terms of computational efficiency of D statistic estimation.

Dataset	Software	Options	Peak memory	Run time
Malawi scaffold_0	Dsuite Dtrios	--no-f4-ratio	92MB	74m59s
	Admixtools qpDstat	blgsize: 0.01	27,212MB	125m2s
	HyDe run_hyde.py	none	178MB	231m38s
	Comp-D†	-d -H -b10	8,300MB	24hours+
	PopGenome	do.df=F block.size=1000	1,170MB	24hours+
Simulation small (20 species)	Dsuite Dtrios	--no-f4-ratio	8MB	28m18s
	Admixtools qpDstat	blgsize: 0.01	17,100MB	13m59s
	HyDe run_hyde.py	none	258MB	19m38s
	Comp-D†	-d -H -b10	22,100MB	24hours+
	PopGenome	do.df=F block.size=1000	440MB	1m50s
Simulation large (100 species)	Dsuite Dtrios	--no-f4-ratio	223MB	215m52s (×100‡)
	Admixtools qpDstat	blgsize: 0.05	1,117,314MB	331m39s (×100‡)
	HyDe run_hyde.py	none	18,716MB	576m32s (×100‡)
	Comp-D†	-d -H -b10	1,000,185MB+	24hours+ (×100‡)
	PopGenome	do.df=F block.size=1000	470MB	274m53s (×100‡)

^†Comp-D cannot use allele frequencies calculated across multiple individuals, so only one individual per species included.

^‡Because of the size of the dataset, we divided the analysis into 100 equally sized jobs to run in parallel; the run time and memory requirements are given for the first job