Table 2. Comparative performance of available orthology calling methods versus RefOGs (32).

					RefOGs
Method	Num. of OGs (RefOGs=67)	RefOGs with F1 ≥85%	RefOGs with Presicion ≥85%	RefOGs with Recall ≥85%	Sum: Exact, Akin	Sum: Fused(events), Split(events)	Sum: Complex, Missed
OrthoDB v8 (2014)	112	51	67	46	43: 30, 13	45: 0(0), 20(45)	4: 4, 0
OrthoDB v5* (2010)	156	42	67	34	33: 24, 9	89: 0(0), 30(89)	4: 4, 0
OrthoMCL (2.0.8)	124	45	64	49	40: 30, 10	51: 2(1), 20(58)	5: 4, 1
COGsoft (4.2.3)	164	29	66	19	19: 12, 7	64: 0(0), 28(64)	20: 19, 1
OMA (0.99t)	224	20	66	13	12: 8, 4	134: 0(0), 31(134)	24: 23, 1

* Used in prior benchmarking (32).

F1 is a harmonic mean of precision and recall (http://en.wikipedia.org/wiki/Sensitivity_and_specificity). RefOG events are defined as follows: ‘Exact’–having 100% of both precision and recall; ‘Akin’–having precision and recall >85% (i.e. up to 1 ‘wrong’ gene for 37% of RefOGs and up to 2 ‘wrong’ genes for another 20% of RefOGs); ‘Fused’–counting fusing events when more than one RefOG represented one method cluster with RefOG recall >85% and summed method cluster precision >85%; ‘Split’–defined symmetrically to Fused when one RefOG is represented by more than one method cluster; ‘Complex’–when the matches can not be classified into another category; ‘Missed’–when a RefOG recall <50%.