Selecting informative subsets of sparse supermatrices increases the chance to find correct trees

BMC Bioinformatics

Table 1 Summary of simulation results

Simulation	Saturation	tic^∗	taxa	Genes	d_QS-value	d _QS	f
						(min/max)	( *correct* ‡)
Gaussian Set1
Unreduced	0.29	0.15	50	50	0.003	(0.99/1.0)	0.01
mare with B^∗	0.69	0.62	9	6	0.0	(0.73/1.0)	0.67
mare with B	0.74	0.74	7	9	0.0	(0.6/1.0)	0.47
Gaussian Set2
Unreduced	0.29	0.1	50	50	0.003	(0.98/0.99)	0
mare with B^∗	0.67	0.61	10	5	0	(0.6/1.0)	0.51
mare with B	0.73	0.73	7	9	0	(0.2/1)	0.42
Power-law non-random Set1
Unreduced	0.13	0.06	50	50	0.17	(0.48/0.99)	0
mare with B^∗	0.46	0.38	25	12	0.02	(0.81/1.0)	0.15
mare with B	0.51	0.51	15	24	0.02	(0.48/1.0)	0.16
Power-law non-random Set2
Unreduced	0.13	0.05	50	50	0.15	(0.43/0.99)	0
mare with B^∗	0.45	0.38	24.5	10	0.06	(0.64/1.0)	0.09
mare with B	0.53	0.53	23	16	0.01	(0.47/1.0)	0.12
Gene threshold Set1
With B^∗	0.72	0.50	34	2	0.05	(0.00/0.42)	0.06
Gene threshold Set2
With B	0.64	0.28	44	3	0.03	(0.00/0.59)	0.03
Gene/taxa threshold Set1
With B^∗	0.59	0.37	21	4	0.05	(0.00/0.46)	0.12
Gene/taxa threshold Set2
With B	0.66	0.30	21.5	4	0.01	(0.00/0.45)	0.25

All values are medians of 100 simulations.
∗ total information content (tic) of un-weighted matrices is allways higher due to the fact that all genes are coded as present/absent (1/0).
‡ f(correct) refers to the frequency of correct trees per 100 simulations.

ISSN: 1471-2105