Efficient context-dependent model building based on clustering posterior distributions for non-coding sequences

Table 4 Number of discrete rate classes when assuming varying rates across sites and the resulting increase in model fit obtained.

Model	Contexts	Annealing	Melting	log BF
GTR+G4	1 (7)	[334.8; 348.4]	[362.5; 376.7]	355.6
GTR+G5	1 (7)	[334.1; 348.1]	[360.4; 374.7]	354.3
GTR+G6	1 (7)	[333.6; 347.8]	[361.3; 375.7]	354.6
GTR+G7	1 (7)	[332.9; 347.1]	[361.6; 375.9]	354.4
GTR+G8	1 (7)	[334.0; 348.0]	[363.7; 378.3]	356.0
GTR	1 (6)	-	-	0

Starting from the default number of four rate classes for the discrete gamma approximation (GTR+G4), we have tested increasing numbers of rate classes. A minor improvement in model fit can be obtained by allowing for eight rate classes (GTR+G8), but such a small improvement may be due to discretization and sampling errors.

ISSN: 2730-7182