Re: [raxml] Digest for raxml@googlegroups.com - 1 update in 1 topic

[Heiko Schmidt]

Dear Chase,

since a pattern is defined as a unique character setting of a column in the alignment, a column containing only gaps has the pattern of “all gaps”, even though that pattern does not contain any information whatsoever.

The two programs just differ in the ways they output this.
While RAxML decides to drop this site and does not even report it as a pattern, IQ-Tree counts it as a distinct pattern.

Both ways are perfectly fine - it might be good if that behaviour would be stated in the respective manuals, I did not check if they do.

Since a gap only column/pattern carries absolutely no information that impacts the likelihood computation, the way of reporting this does not have any consequences and makes no difference to the ML tree search.

However, one could discuss is: in the case that the all-gaps pattern is not counted and maybe excluded from the ML computation, should then the reported length of the alignment be reduced by the number of sites holding the all-gaps pattern?

In my opinion, the alignment should not contain any all-gap columns at all, because they are typically an artefact caused by removing sequences from a larger alignment (or from manually correctly the alignment).
However, it would be unpleasant if the ML software would through am error and refuse to run if it detects such sites, because in the ML computation such patterns do not change the resulting tree.

Best regards,
Heiko


> On 10.03.2023, at 13:00, ra...@googlegroups.com wrote:
>
> ra...@googlegroups.com Google Groups
> Topic digest
> View all topics
> • Number of patterns: raxml-ng vs. iqtree - 1 Update
> Number of patterns: raxml-ng vs. iqtree
> "Chase W. Nelson" <cwnel...@gmail.com>: Mar 09 09:51PM -0800
>
> I've been noticing that raxml-ng consistently reports *one fewer pattern*
> than does IQTREE for the same nucleotide alignment. For example:
>
> *raxml-ng:*
>
> - Loaded alignment with 5837 taxa and 7906 sites
> - WARNING: Fully undetermined columns found: 319
> - Alignment comprises 1 partitions and *5882 patterns*
> - ...
> - Alignment sites / patterns: 7587 / 5882
>
>
> *iqtree:*
>
> - WARNING: 319 sites contain only gaps or ambiguous characters.
> - Alignment has 5837 sequences with 7906 columns, *5883 distinct
> patterns*
> - 1545 parsimony-informative, 1360 singleton sites, 5001 constant sites
>
>
> Has anyone else experienced this or found the reason for the one-off
> difference? (Perhaps iqtree is counting a fully undetermined column as a
> pattern — but if so, why, and does that have any consequence?)
>
> Thanks a ton!
> Chase
> Back to top
> You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page.
> To unsubscribe from this group and stop receiving emails from it send an email to raxml+un...@googlegroups.com.

-----------------------------------------------------------------------------
Heiko Schmidt
Vienna, Austria
-----------------------------------------------------------------------------