Unknown RAxML options: Fast and superficial tree searches

[Alexis]

Dear All,

I just wanted to pint your attention to the fast and superficial tree
search option "-f E" that has been available in RAxML for quite some
time now, but rarely used.

In terms of execution times it ranges in the same order of magnitude
as FastTree, but usually yields trees with significantly better
likelihood scores.

Cheers,

Alexis

[Ingo Michalak]

Dear Alexis,
thanks for pointing to this.

We'd like to make it accessible in the next version of raxmlGUI.
Actually we thought of adding an optional pipeline consisting of this
fast tree search, and successive calculations of branch lengths (RAxML
option "-f e") and SH-like support values (RAxML option "-f J"), so
users could get quite fast a tree with branch lengths and node
supports (according to our first tests a rough estimate is in about
1/5 of the time of a simple ML search "-f d"). My question would be if
there is any conceptual issue with this combination: would such a tree
be comparable to the FastTree output?

Thanks in advance and best regards

Ingo

[Alexis]

Dear Ingo,

I believe that such an option makes sense. In general I would expect the trees to have a better log likelihood score than those obtained via FastTree. Nonetheless, I'd include a user warning that this will just execute a fast and relatively superficial tree searches, hence caution is advised.

Cheers,

Alexis

[SineadH]

Hi,

I'm running a very large dataset (14000~ taxa) using the -f E option
and pthreads -T 3.
It has been running for over 300 hours now.

I was wondering do you know roughly how long it will take?
The largest dataset I could get to run with FastTree was 10000.
I'm hoping this one isn't too big! :)

Thanks!

Sinéad.

[Fernando Izquierdo]

Hi Sinead,

It is difficult to predict the running time, but you can have a look
at the RAxML_log file to get an idea of how close you are getting to
convergence.
Is this a protein or a dna dataset? How large is it in terms of
characters/partitions?
Also make sure you are using the latest standard-RAxML version from
the github repo, since the convergence criterion for -f E was recently
changed.

Cheers,
Fernando

[SineadH]

Hi,

It's a protein dataset, 1301 characters long.
It doesn't seem to be producing a log file but the info file is
printing what I thought was likelihoods.
Do you think it's probably not running properly?

I think I'm using an older version, I didn't download it from the
github.
I'll get the newest version and try that.

Thanks!!

Sinéad.

On Mar 20, 10:58 am, Fernando Izquierdo <fer.izquie...@gmail.com>
wrote:

> Hi Sinead,
>
> It is difficult to predict the running time, but you can have a look
> at the RAxML_log file to get an idea of how close you are getting to
> convergence.
> Is this a protein or a dna dataset? How large is it in terms of
> characters/partitions?
> Also make sure you are using the latest standard-RAxML version from
> the github repo, since the convergence criterion for -f E was recently
> changed.
>
> Cheers,
> Fernando
>