HyPhy message board - BranchSite REL and Bayes Factors

	Welcome, Guest. Please Login

Home

Help

HyPhy message board › HYPHY Package › HyPhy feedback › BranchSite REL and Bayes Factors

(Moderators: Sergei, Simon)

‹ Previous Topic | Next Topic ›

Pages: 1

Send Topic

BranchSite REL and Bayes Factors (Read 3551 times)

fabia

YaBB Newbies

Offline

Feed your monkey!

Posts: 8

BranchSite REL and Bayes Factors
May 25^th, 2011 at 4:32pm

Dear all,

I am using the REL method to estimate the omega for each site of my alignment. I followed the tutorial on the InfluenzaA_H3_Random35.nex file that outlines two steps:
1. run dNdSRateAnalysis.bf to evaluate if synonymous rates vary from codon to codon
2. run dNdSResultProcessor.bf to find which sites are under positive selection.

The threshold given in the exercise for positive selection is a Bayes Factor of 50 but I was not able to find a good explanation for this number. Is it data set specific? Obviously, lowering this number will give larger number of positively selected sites so I think it is a rather important parameter and I would like to know the reasoning to apply to choose an appropriate value.

Thanks,
Fabia

IP Logged

Sergei

YaBB Administrator

Offline

Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male

Re: BranchSite REL and Bayes Factors
Reply #1 - May 27^th, 2011 at 10:20am

Hi Fabia,

I tend to think of Bayes Factors (in this setting) as the reciprocals of p-values. 1/50 ~ p-value 0.02. The most sensible way to choose a BF (or a p-value) is to confirm that the test has a proper size, and then decide what your tolerance to false positives is. There is no compelling reason to use p=0.05 or p=0.01 -- it's just a matter of convention.

Sergei

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego

WWW

IP Logged

fabia YaBB Newbies Offline Feed your monkey! Posts: 8	Re: BranchSite REL and Bayes Factors Reply #2 - May 27^th, 2011 at 10:31am Hi Sergei, thanks for the answer! I have another very quick question: in HyPhy I can use branch models (global or local) or site models to estimate the omega. If I use a site model with local parameters will that be similar to a branch-site model (like the ones implemented in PamL)? Fabia
Back to top	IP Logged

Sergei YaBB Administrator Offline Datamonkeys are forever... Posts: 1658 UCSD Gender:	Re: BranchSite REL and Bayes Factors Reply #3 - May 27^th, 2011 at 10:33am Hi Fabia, No -- for the analog of branch-site models, please use Branch-Site REL (Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to Login) Sergei
Back to top	Associate Professor Division of Infectious Diseases Division of Biomedical Informatics School of Medicine University of California San Diego WWW IP Logged

fabia YaBB Newbies Offline Feed your monkey! Posts: 8	Re: BranchSite REL and Bayes Factors Reply #4 - May 27^th, 2011 at 10:38am Perfect, I will try that. Thanks! Fabia
Back to top	IP Logged

fabia

YaBB Newbies

Offline

Feed your monkey!

Posts: 8

Re: BranchSite REL and Bayes Factors
Reply #5 - May 27^th, 2011 at 11:45am

Hi Sergei,

I did a test run with the BranchSiteRel.bf file that you suggested. I have three questions:

1. this method is supposed to identify the branches under episodic selection. Is it the same as saying branches under positive selection?

2. the csv file produced lists the percentage of site that are at a given omega for each branch. So, if I have this output:

Branch Mean_dNdS Omega1 P1 Omega2 P2 Omega3 P3 LRT p p_Holm
SRC1 0.382373 0.380406 0.958053 0.623567 2.44E-08 0.630251 0.0419473 0 1 1

Am I correct in saying that for branch SRC1 96% of the sites have an omega of 0.38, 0.0000024% omega=0.62, and 4% omega=0.63? In this case, omega2 and omega3 are very similar, why are they separated in different categories?

3. Is there a way to get the list of sites under positive selection?

Thanks again,
fabia

IP Logged

Sergei

YaBB Administrator

Offline

Datamonkeys are forever...

Posts: 1658
UCSD
Gender: male

Re: BranchSite REL and Bayes Factors
Reply #6 - May 27^th, 2011 at 1:04pm

Hi Fabia,

1). By definition, episodic selection in this context means selection along a subset of branches, so you are correct; the method looks for evidence of positive selection along individual branches at a non-zero proportion of sites.

2). Your interpretation is correct. The number of categories is selected a priori (3), and it your particular case that's too many -- you effectively have 2 rate classes. Note that the 2nd rate class has no sites in it (P2 ~ 10^-8).

3). We have a separate method for that (in preparation). The sample size for inferring selection along one site along one branch is 1, hence we can either infer branches where SOME sites have undergone selection (BranchSiteREL) or sites where SOME branches have undergone selection -- not both.

HTH,
Sergei

Associate Professor
Division of Infectious Diseases
Division of Biomedical Informatics
School of Medicine
University of California San Diego

WWW

IP Logged

Pages: 1

Send Topic

‹ Previous Topic | Next Topic ›

« Home

‹ Board

Top of this page