HyPhy message board | |
http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl
Methodology Questions >> How to >> Likelihood value for constrained Q matrix http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?num=1269002688 Message started by Miguel Lacerda on Mar 19th, 2010 at 5:44am |
Title: Re: Likelihood value for constrained Q matrix Post by Sergei on Mar 20th, 2010 at 9:20am
Hi Miguel,
If you are using a HyPhy build that is not than ~8 months old, then the calculation for the mixed (along) branches model should be correct. That said, it is possible that by adding a new REL model to the subset of branches can decrease the likelihood of a single REL model. Consider the calculation of single site likelihood at a site under a two omega REL model, i.e. Code (] Pr[omega = A):
Hence, Code (] L[site):
Now consider splitting the branches in a tree into two (arbitrary, non-empty) classes, and assigning a separate copy of the same REL model to both classes (hence creating 4 rate classes). This seems like what you are doing with models C and V initially. The likelihood calculation is now Code (] L[site):
Notice that the new likelihood can be LOWER than the previous one, because the two terms corresponding to the single model likelihood have weights adding to LESS than one (p^2 + (1-p)^2), and the other two likelihood terms could DECREASE the likelihood. HTH, Sergei |
Title: Re: Likelihood value for constrained Q matrix Post by Miguel Lacerda on Mar 21st, 2010 at 3:23am
Makes sense! Thanks... Miguel
|
HyPhy message board » Powered by YaBB 2.5.2! YaBB Forum Software © 2000-2024. All Rights Reserved. |