RAxML-EPA -G option

[Alicia S. Arroyo]

Hello all, 

I have a quick question about EPA, but I couldn't find a proper answer on this google group. 

I usually call EPA with this command, as an example: 

raxmlHPC-PTHREADS -f v -G 0.2 -m GTRCATI -n EPA -s file.fasta -t RAxML_bestree.tre -T 6

I tried to figure out was -G option means. The RAxML manual says " -G: enable the ML-based evolutionary placement algorithm heuristics by specifiying a threshold value (fraction of insertions branches to be evaluated using slow insertions under ML)"

So, does this threshold value, 0.2 in this case, mean that only 20% of the insertions of each short read are considered? Because if this is the case, I would like this value to be 1 because I want a evaluation of all the insertions, right? I can't understand why someone wouldn't like to consider all the posible insertions of the short sequences in the reference tree. 

But also, this sentence could be also understood as: only 20% of the reference branches in the reference tree are considered... but this doesn't really make sense. 

Any more complete explanation for this sentence, please?

Thank you very much!

A. 

[Alexey Kozlov]

Hi Alicia,

so first of all, please note that "-G" option is a mere performance optimization - so on small datasets (and small
reference trees in particular) you can safely disable it.

This optimization works as follows:

- In the first "pre-scoring" step, all queries are inserted in all branches of the reference tree *without* optimizing
the branch lengths.

- Then, only 20% (-G 0.2) of the best-scoring branches *for each query* are re-evaluated more thoroughly, by doing
re-insertions with branch length optimization.

This saves quite some time, since optimizing the branches is rather expensive compared to the simple likelihood evaluation.

Hope I clarified it a bit...

Cheers,
Alexey

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

[Alicia S. Arroyo]

Hi Alexey, 

Thank you very much, this is exactly what I needed :) Clear now! 

Thanks!!!!

A.