Hello,

The results (i.e. with regard to the number of lineages under the influence of diversifying selection) obtained, for the same dataset, from the BS-REL analyses  in the current versions of Hyphy (version 2.2) and the Datamonkey differ very much from each other. However, the outputs from the BS-REL analyses in the older version of the HYPHY package (version 2.1.2) is the same as that obtained from the current version of the Datamonkey. Just wondering what I am doing wrong or what is going on?

Thank you very much in advance for your input!

Best,

Flu

Hi Flu,

The version of BS-REL implemented in version 2.2 of HyPhy is quite different from 2.1.2 and the Datamonkey version. Whereas the old version used 3 dN/dS classes per branch, the new model adaptively chooses the number of rate classes, resulting in a simpler model. How different were your results? Did the 2.2 version detect more branches under selection?

Sergei

Hi Sergei,

Thank you for the response. The answer to your question is MORE (i.e. 3 more branches appear to be affected by positive selection when I ran the data in Hyphy), that is when when I answered NO to the following question: "Allow Branch-site variation in synonymous rates". When I chose YES as the answer to the above question the differences in the results were more prominent. However, I am rerunning the analysis again to verify that result .

Thanks again for your input.

Best,

Flu

Hi Sergei,

I have now obtained the results from rerunning the data set in HYphy (BS-REL  program) when "allowing Branch-site variation in synonymous rates" and the outcome was that NONE of the branches were affected by episodic diversifying selection.

For the same dataset the following results are obtained with regard to the number of branches affected by episodic diversifying selection:

1. Datamonkey (BS-Rel program)  15 branches.

2. Hyphy (BS-Rel program; NOT allowing Branch-site variation in synonymous rates)      18 branches

3. Hyphy (BS-Rel program;  ALLOWING Branch-site variation in synonymous rates)        0 branches

Which of these 3 approaches is preferred? Does situation 3 mean lack of power and situations 1 and 2 many false positives or...? I am not sure how to interpret the above results. I expect situation 3 to differ from situations 1 and 2 but not as drastically as the outcome is in this particular analysis.  I very much appreciate your input and I can send you the data set if you want to look at it further?

Thank you very much in advance for your input.

Best,

Flu

Hi Flu,

Option (3) is not really meant to do selection testing: we are currently developing it to improve the estimation of the evolutionary process (branch lengths etc), so I would not worry about that.

It is expected that (2) will have better power than (1), because in (2) we use models with fewer parameters, improving numerical stability and power to detect selection. Currently, option (2) would be my suggested default.

Sergei

Thanks Sergei for your suggestion.

Best,

Flu