HyPhy message board | |
http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl
Theoretical questions >> Sequence Analysis >> When to trust GA analysis http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?num=1161792492 Message started by Sarah on Oct 25th, 2006 at 9:08am |
Title: When to trust GA analysis Post by Sarah on Oct 25th, 2006 at 9:08am
I have a broad question.
Under what conditions can we trust the results of the GA branch selection analysis? With PAML one can infer whether nested hypotheses are significant improvements over one another, and there are also indications (e.g., whether the dN/dS value is insanely high) if the inferences are weak. Are there any papers on this topic? Could you spell out what to look for, or list cases in which GA should perform poorly? Thanks. Sarah |
Title: Re: When to trust GA analysis Post by Sarah on Oct 25th, 2006 at 10:40am
Yes, 'trust' is too... trusting. I suppose I'm looking for more indicators of confidence, and I don't have much experience with ML (and obviously didn't write the algorithm!).
I'm going to check out the book you recommended. Thanks for listing the diagnostics--they are helpful. Sarah |
Title: Re: When to trust GA analysis Post by Sergei on Oct 25th, 2006 at 1:08pm
Dear Sarah,
Good luck! I like this quote from Edwards' 'Likelihood' All our likelihood arguments are conditional on particular probability models: in a sense the model itself is a nuisance parameter. We would like to argue without it, but cannot. Isn't statistical inference fun? I was actually going to modify the web implementation of GA Branch to run faster, use a tweaked (from our more recent papers) algorithm for the search, and automatically determined the appropriate number of rate classes. Stay tuned... Cheers, Sergei |
Title: Re: When to trust GA analysis Post by Sergei on Nov 1st, 2006 at 7:04pm
Dear Sarah,
I just finished updating the GA Branch back end scripts; they will now run about 10-50 times faster depending on the data, and also automatically cycle through the number of rate classes, from 2 up to a maximum of 10. Cheers, Sergei |
Title: Re: When to trust GA analysis Post by Sarah on Nov 1st, 2006 at 7:06pm
Awesome! Thanks! I look forward to trying it out. ;D
Sarah |
HyPhy message board » Powered by YaBB 2.5.2! YaBB Forum Software © 2000-2024. All Rights Reserved. |