sCF values between 0 and 33% for a strong dataset

[Dominic Evangelista]

Greetings,

I understand that sCF values of 33% indicate that all quartets are equally likely around a node and sCF of 0 means there are no parsimony informative sites for that relationship. However, what does it mean if they are less than 33% but more than 0?

In principal I don't understand why this would happen. I have attached my .stats and .log file. I used -scfl 200

But further than that, I am not sure why these values would be low for my particular dataset. I have a 35 taxon tree. There are 75,825 parsimony informative sites and the alignment should have very high signal for these data. FcLM map for the whole data (n = 50,000 random quartets) show 0% unresolved nodes. Yet, the mean sCF value is 38 and the lowest values are 26-33%.

I can share the alignment and tree privately if needed.

[Dominic Evangelista]

Thinking about this more, I suppose the answer is obvious: if sCF<33% most of the sites in the alignment support the alternative topologies. In other words, this is not the best tree.

[Dominic Evangelista]

Having answered my own first question I am more confused about my second question. Tree attached (alignment wouldn't attach)

For example: look at a single quartet on my tree

(((Blaberus, Schultesia), Nauphoeta), Remaining)

That quartet is recovered in all 10 full tree inferences (with model finder etc), and has 100% bootstrap but only 26% sCF. sDF1 and 2 suggest that the alternatives have thousands more sites supporting them. Not only does this not make sense given prior knowledge of the phylogeny, but it also doesn't make sense given the analyses.

Is it possible there is a problem with the model specification? Or...?

[Dominic Evangelista]

Sorry for so many replies (wish I could edit my previous post). I should also add, that between inferring the tree and calculating sCF values I rerooted my tree. I don't see how that would affect anything but thought it was important to mention anyway.

[Dominic Evangelista]

Ok I think this is an error with the -scfl option. I redid it with the older -scf option and the numbers make MUCH more sense.

[rob.l...@gmail.com]

Hi There,

Thanks so much for this post. If you are getting drastically different numbers between --scf and --scfl it could be due to a bug in the scfl calculations. 

If you could open a GitHub Issue (https://github.com/iqtree/iqtree2/issues) for this case, and share your alignment, ML tree, log files from the ML tree analysis, and input files (either on the issue, or us a link via email if you prefer not to upload them), we'd be really grateful. 

There's a chance that this is a genuine difference between what likelihood says and what parsimony says too. But from the other info you've given, a bug is certainly high on the list of possible answers!

Many thanks,

Rob

[rob.l...@gmail.com]

P.S. Always nice to see a phylogeny of cockroaches!