Table 2.

Computational efficiencies on the OrthoMaM data set for the tested methods

	Concat+Dist	Concat+ML	SDM*add	DistRadd	ERaBLEadd	SDM*	DistR	ERaBLE
T 1	≈0	3 h 20 m/39 h 28 m				2 m 46 s	2 m 46 s	2 m 46 s
T 2	5 m 41 s	41 h 16 m	8 h 2 m	2 h 9 m	7 s	8 h 33 m	2 h 6 m	7 s
M	889 MB	117 GB	1.2 GB	2.8 GB	222 MB	1.2 GB	3.0 GB	221 MB

Note.— The first row gives (T ₁) the running time to obtain the data on which subsequent computations are based: the superalignment and the distance-based gene trees for Concat+Dist, the superalignment and ML gene trees for Concat+ML, the ML gene trees and resulting additive distances for the three supertree methods, and the estimated distances for the three medium-level methods. When ML gene trees are used (Concat+ML, SDM*add, DistRadd and ERaBLEadd), two alternative approaches are possible and therefore two running times are provided: first that to infer trees with fixed topology (3 h 20 m), and then that to infer trees where the topology is also estimated (39 h 28 m). The second row gives (T ₂) the remaining running time to obtain estimates for branch lengths and gene rates. The third row (M) gives the maximum amount of memory allocated. All the experiments were conducted on a PC with 4 GB RAM and a 2.7 GHz CPU, except branch length estimation (T ₂ and M) for Concat+ML, which, because of the large memory requirements, was run on a cluster machine with 200 GB RAM and a 2.66 GHz CPU