Implementing Jackknifing

[Emmanuel Toussaint]

Hello,

Are you thinking of implementing the jackknife as a mesure of support in addition to the bootstrap and UFBoot? I believe the use of jackknifing would be an interesting complement to assess topological robustness and it would certainly be interesting to be able to estimate jackknife support for phylogenomic datasets in IQ-TREE.

Best,

Emmanuel

[Bui Quang Minh]

Hi Emmanuel,

This is a reasonable request. I will add this to the TODO list as I don’t see any difficulty to implement jackknife. 

Just one question: how many sites will one typically subsample (without replacement) in a jackknife analysis?

Minh

--
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

[Karen Meusemann]

Hi Minh,

if you implement this, would you in addition also (recommend)implement taxon jacknifing?

Best, Karen

[Emmanuel Toussaint]

Hi Minh,

Thanks for the quick reply, I think this will be useful. I don't think there is a consensus on the jackknife rate to use but based on several studies on the topic I think 40% is a good starting point.

Emmanuel 

[Bui Quang Minh]

Hi Emmanuel,

Then the best is some option like -j 0.4 to specify jackknife rate of 40%.

@Karen: taxon jackknife can be implemented, but the final step of consensus tree does not work as trees do not have the same taxon set. Then you have to use some other supertree method to summarize the trees.

Minh

[Celine Brochier-Armanet]

Hi,

Thanks for implementing the jackknife procedure. This is very useful, especially when working on very large multiple alignments.

I tried to use it recently with the command line :

iqtree -s alignment.phy -j 0.1 -m GTR+F+R4

The ML tree was accurately inferred (using all positions of the alignment), but the jackknife procedure did not start (and no supports are associated to branches).

I am using IQ-TREE multicore version 1.6.12 for Linux 64-bit built (Aug 15 2019).

One more question, how many jackknife replicates are performed with the default parameters ? Is it possible to choose the number of replicates ?

Many thanks in advance

Best wishes

Céline Brochier

[Minh Bui]

Hi Celine,

There are differences between IQ-TREE 1.X and 2.0 for the jackknife option (will be clarified soon in the documentation):

For v1.X, you need to specify the jackknifing proportion via -j option and the number of replicates either via -b or -bb. For example, 

"-j 0.1 -b 100" will invoke the standard jackknife with 100 replicates and 0.1 jackknife proprotion.

“-j 0.1 -bb 1000” will invoke 1000 ultrafast jackknife, in the same manner as the ultrafast bootstrap. So this will be much faster, but the jackknife support values might have a different interpretation. However, we didn't examine this: one can perform a simulation study as in our paper (https://academic.oup.com/mbe/article/30/5/1188/997508) to investigate its behaviour compared with the “standard” jackknife.

So “-j 0.1” alone does not have any effect, as you didn’t specify the number of replicates.

For v2.0, we change the semantic of the -j option, which is to specify the number of replicates, where the default jackknife proportion is 0.5 (there was a paper claiming that this is the best parameter, but I can’t recall right now). This can be changed via --jack-prop option. So for example,

“-j 100” will perform the standard jackknife with 100 replicates and jackknifing proprotion 0.5.

“-J 1000” will perform the ultrafast jackknife with 1000 replicates and jackknifing proprotion 0.5.

“-j 100 --jack-prop 0.1” will perform the standard jackknife with 100 replicates and jackknifing proprotion 0.1.

“-J 1000 --jack-prop 0.1” will perform the ultrafast jackknife with 1000 replicates and jackknifing proprotion 0.1.

Hope that helps,

Minh

--
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 view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/ddee439e-ef35-49b9-b6ca-7d886297628e%40googlegroups.com.

[Celine Brochier-Armanet]

Many thanks

Best wishes

Céline

Le mardi 3 décembre 2019 02:08:30 UTC+1, Minh Bui a écrit :

Hi Celine,

There are differences between IQ-TREE 1.X and 2.0 for the jackknife option (will be clarified soon in the documentation):

For v1.X, you need to specify the jackknifing proportion via -j option and the number of replicates either via -b or -bb. For example, 

"-j 0.1 -b 100" will invoke the standard jackknife with 100 replicates and 0.1 jackknife proprotion.

“-j 0.1 -bb 1000” will invoke 1000 ultrafast jackknife, in the same manner as the ultrafast bootstrap. So this will be much faster, but the jackknife support values might have a different interpretation. However, we didn't examine this: one can perform a simulation study as in our paper (https://academic.oup.com/mbe/article/30/5/1188/997508) to investigate its behaviour compared with the “standard” jackknife.

So “-j 0.1” alone does not have any effect, as you didn’t specify the number of replicates.

For v2.0, we change the semantic of the -j option, which is to specify the number of replicates, where the default jackknife proportion is 0.5 (there was a paper claiming that this is the best parameter, but I can’t recall right now). This can be changed via --jack-prop option. So for example,

“-j 100” will perform the standard jackknife with 100 replicates and jackknifing proprotion 0.5.

“-J 1000” will perform the ultrafast jackknife with 1000 replicates and jackknifing proprotion 0.5.

“-j 100 --jack-prop 0.1” will perform the standard jackknife with 100 replicates and jackknifing proprotion 0.1.

“-J 1000 --jack-prop 0.1” will perform the ultrafast jackknife with 1000 replicates and jackknifing proprotion 0.1.

Hope that helps,

Minh

On 3 Dec 2019, at 3:46 am, Celine Brochier-Armanet <celine.bro...@gmail.com> wrote:

Hi,

Thanks for implementing the jackknife procedure. This is very useful, especially when working on very large multiple alignments.

I tried to use it recently with the command line :

iqtree -s alignment.phy -j 0.1 -m GTR+F+R4

The ML tree was accurately inferred (using all positions of the alignment), but the jackknife procedure did not start (and no supports are associated to branches).

I am using IQ-TREE multicore version 1.6.12 for Linux 64-bit built (Aug 15 2019).

One more question, how many jackknife replicates are performed with the default parameters ? Is it possible to choose the number of replicates ?

Many thanks in advance

Best wishes

Céline Brochier

--
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 iqt...@googlegroups.com.