Tree Topology tests c-ELW values

[aditya kumar]

Hello,

I would like to know more details regarding c-ELW values and the calculation of confidence set of trees. If there were some trees with very low ELW values, which tree is rejected and how is the ranking order defined.

As an example if 4 tree topologies are tested and the elw values are as follows
T1 is 0.8523
T2 is 0.0303
T3 is 0.0871
T4 is 0.0303

which tree would be rejected, T2 or T4, if both of them have same likelihood values while T1 is the best tree with highest likelihood.

 How is the decreasing order defined if two trees have same elw values and if the corresponding likelihood values are also same.How IQ-Tree assigns the "-" sign for tree models with same elw values.

Thank you.

[Heiko Schmidt]

Dear Aditya,

> I would like to know more details regarding c-ELW values and the calculation of confidence set of trees. If there were some trees with very low ELW values, which tree is rejected and how is the ranking order defined.

To explain the whole method how would take too long here. I would like to refer you either to the ELW paper or the “Testing Tree Topologies” in “The Phylogenetic Handbook” where I tried to explain it graphically.

However, using randomisation ever tree gets a weight (where the weights of all trees add up to 1.0).

Then the trees are ordered by their weights. Although some weights might look identical, in reality they rarely are (unless you used identical trees). So the printed (and thus rounded) weights may differ beyond the printed precision.

Then you collect all trees until the summed-up weights reach 95%, and all those which are in are called the 95% confidence set.

> As an example if 4 tree topologies are tested and the elw values are as follows
> T1 is 0.8523
> T2 is 0.0303
> T3 is 0.0871
> T4 is 0.0303
>
> which tree would be rejected, T2 or T4, if both of them have same likelihood values while T1 is the best tree with highest likelihood.

If the two really have the same likelihood and weight, I would rather check if the tree topologies are identical, because this should not happen and then remove all but one of them. A second possibility might be that you have identical sequences (or such that only differ by wildcards) in there, and than the programme (actually no programme) can tell the difference in likelihood. If these occur on the boundary of the confidence set, it might be better to remove the identical sequence except one and work with the reduced dataset. Identical sequences will end up anyway on the same subtree separated by near-zero branch lengths.

> How is the decreasing order defined if two trees have same elw values and if the corresponding likelihood values are also same.How IQ-Tree assigns the "-" sign for tree models with same elw values.

As mentioned above, the question is whether the values of the trees are really identical. And even if you have several trees with (almost) identical ELWs around the cutoff, one has to treat them with care anyway.
In the ideal case you have very few trees with pretty high ELWs and the low ones don’t even get close to the cutoff level.

I hope that helps a bit.

Best regards,
Heiko

[aditya kumar]

Dear Heiko,

Thank you very much for the quick response and detailed explanation for all my questions.

Actually I am interested in the trees which are not falling into the cut-off values. So, order of the elw values would affect which trees would be rejected. Is there a possibility where I can see the precision of the ELW values to make a clear inference.

thank you,

Best Regards,

Aditya

[Bui Quang Minh]

Hi Aditya,

On Jan 9, 2018, at 7:27 PM, aditya kumar <adina...@gmail.com> wrote:

Dear Heiko,

Thank you very much for the quick response and detailed explanation for all my questions.

Actually I am interested in the trees which are not falling into the cut-off values. So, order of the elw values would affect which trees would be rejected.

Yes the order will affect which trees being rejected, but only for the one tree at the margin of the 95% confidence set.

Is there a possibility where I can see the precision of the ELW values to make a clear inference.

No that’s not possible unfortunately. The precision of 4 is hard-coded when writing .iqtree file.

Minh

thank you,

Best Regards,

Aditya

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

--

Bui Quang Minh

Center for Integrative Bioinformatics Vienna (CIBIV)

Campus Vienna Biocenter 5, VBC5, Ebene 1

A-1030 Vienna, Austria

Phone: ++43 1 4277 74326

Email: minh.bui (AT) univie.ac.at

[aditya kumar]

Dear Minh,

Thank you very much.

Best Regards,

Aditya