Diagnostics using the Effective sample sizes (ESS) and Stabilisation

[W. Yin]

Hi, BAli-Phy author and users,

I'm not a bioinfo nor math guy, but just a wet-lab biologist. I hope I could have some straightforward answers from you, if that's possible, so that I could understand. 

I was struggling quite a bit to use the BAli-Phy to make gene trees. After I closely read the User Tutorial and Guide, and thanks to the clarity of these documents, I managed to get some nice and interesting trees, which make more biological sense to me compared with the Clustal-O NJ established trees.

My question is, how to use the ESS and Stabilisation to determine when to stop the program. I know it's probably explained and answered in the section 11 and FAQ of the user guide. but, still, I'm not very confused.

I currently use the criteria ASDSF < 0.01 and PSRF CI80%/RCF < 1.01 to decide when to end the program. but really not sure if this is OK using only two rather than all the diagnostics mentioned in the user guide. and does these diagnostics (ASDSF, PSRF, ESS and stabilisation) have a priority order that I should follow? 

many thanks and best wishes,

W. Yin

[Benjamin Redelings]

Hi,

I'm glad you think the tree estimates are interesting!

I think the obvious thing to do is to check that ASDSF < 0.01 and that the ESS values are large enough.  The ASDSF checks that the tree topologies are converging, and the ESS checks that the parameter estimates are converging.  There is not really a magic number for the ESS, but maybe all the ESS numbers should be > 500.  Probably this gives similar results to looking at the ASDSF and the PSRF.

However, I think the ultimate goal is not to follow some simple rules, but rather to convince yourself that (i) the multiple different chains are converging to a common equilbrium, (ii) the results will not change if you run it longer, and that (iii) if you run the analysis a second time, your results should be similar.

In order to convince yourself that the multiple chains are converging, all the different diagnostics are helpful, but you don't have to use all of them.  If you are in doubt, you could always run the chains a bit longer and see if anything changes, or run the analysis again and see if you get the same results.

Does that help?

-BenRI

--
You received this message because you are subscribed to the Google Groups "bali-phy-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bali-phy-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bali-phy-users/97589541-7b83-4f5a-aa03-1dd139752eaf%40googlegroups.com.

[W. Yin]

Dear BenRI,

Thank you so much for your kind reply and very detailed explanation. Sorry for replying late. when I submitted the question in the first place I couldn't see it in the forum. I thought my question submission was failed. I'm happily surprised today to see my question has been nicely answered. yes, your answers are very helpful. just one more question, which PSRF should I look at? the PSRF-80%CI seems always different/smaller than PSRF-RCF, and I can't really get it the meaning of the 80%CI and RCF.

best wishes,

W. Yin

在 2020年5月26日星期二 UTC+1下午3:09:29，benjamin...@gmail.com写道：

To unsubscribe from this group and stop receiving emails from it, send an email to bali-ph...@googlegroups.com.