Questions about FDR with the results of BUSTED

[SWei2333]

Dear professor,

I performed BUSTED analysis on 7,401 genes to identify convergent positively selected genes. Due to the species I selected for my study, I ended up with very few significant p-values. When I applied FDR correction, many ideal candidate genes were excluded. May I ask if it is acceptable to proceed with the analysis using the raw p-values, or is it better to rely on the q-values obtained after FDR correction?

[SWei2333]

[spond]

Dear @SWei2333,

1. How many sequences do you have? The fraction of significant results looks pretty low.
2. What BUSTED options did you use (and which version did you run)?
3. It looks like you simply may have unavoidable power limitations. Instead of running genome wide screens, if you have a set of candidate genes beforehand, you might want to constrain testing to those genes. Relaxing test stringency is inviting, but if you do that, then you can't control false positives. I would at least continue using FDR.

Best,
Sergei

[SWei2333]

Dear Professor,

1. I have a total of 7,401 genes.

2. My command(HYPHY 2.5.42) :
   hyphy BUSTED --alignment ../one2one.filter.fa/ENSMUST00000000001.4.flt.cds.fa --tree ./one2one.unroot.FG.tree/ENSMUST00000000001.4.unroot.FG.nwk --branches Foreground --output ./result/ENSMUST00000000001.4.json

3.I would like to ask why reducing the number of genes included in the analysis would affect the p-values. Isn’t each gene analyzed independently in HyPhy?

4. I am not sure if the reason for the small number of significant genes is due to my species selection. My gene tree is structured in such a way that it can only be divided into four groups based on traits.（Branches with color are set as foreground branches.）
   

[spond]

Dear @SWei2333,

It does seem like you simply have a small sample situation; relatively few branches which are relatively short. I would suggest

1. Get the latest hyphy version, because I did a lot of work on BUSTED in the last year.
2. For smaller datasets, it sometimes helps to simplify the model

hyphy busted --rates 2 --syn-rates 2 --starting-points 5 --branches Foreground ...

3. If you reduce the number of genes you tested, the p-values themselves don't change, but you have to correct for fewer tests.

Best,
Sergei

[SWei2333]

Dear professor, Thank you for your help! I'll have a try. Best wishes Sergei Pond ***@***.***> 于2024年11月26日周二 01:04写道：

…

Dear @SWei2333 <https://github.com/SWei2333>, It does seem like you simply have a small sample situation; relatively few branches which are relatively short. I would suggest 1. Get the latest hyphy version, because I did a lot of work on BUSTED in the last year. 2. For smaller datasets, it sometimes helps to simplify the model hyphy busted --rates 2 --syn-rates 2 --starting-points 5 --branches Foreground ... 3. If you reduce the number of genes you tested, the p-values themselves don't change, but you have to correct for fewer tests. Best, Sergei — Reply to this email directly, view it on GitHub <#1767 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BIAZCE2TM5OOFJWFEX6B5LT2CNKCLAVCNFSM6AAAAABSN5WWHOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDIOJYGU3DQOJTGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>