How construct balance tree from rooted-tree

I’m trying to follow the Differential abundance analysis with gneiss ( with my own data. I tried the correlation-clustering and gradient-clustering method to construct balance. But they took long time… I run them overnight and haven’t got any output… So I tried to use rooted-tree to continue the ilr-hierarchical step. But it returned
Argument to parameter ‘tree’ is not a subtype of Hierarchy.
I don’t know how to transform the rooted-tree into a Hierarchy type… and I am also confused that whether should I continue to running the correlation-clustering…Is long-time running normal for this step?

Thank you!

Hey @nan_wang,

You need to run ilr-phylogenetic instead of ilr-hierarchical) which accepts phylogenetic trees (and will return a re-oriented hierarchy for further user).

Hope that helps!

Hi Bolyen,

Thank you very much for your reply. I run it on my MacBook pro (2017 version, with 16GB RAM) but it returned ‘Killed: 9’ without any other information. When it is running, I noticed that the load of memory was periodically heavy, and python & kernel_task alternately run. I’m wondering why it happens? is it because of beyond the capacity of memory? If so, how can I continue this analysis?
Thank you very much!


Hi @nan_wang,

Yes memory is probably why it was killed, although I would have expected 16gb would be enough.
@mortonjt, how much memory does ilr-phylogenetic usually take?

That is strange. How many taxa are we talking about?

Hi Morton, Hi Bolyen,
Thank you for your reply.
I’m not sure where to know the taxa, but the feature table shows that
Number of features 61,160
Total frequency 3,269,100
Are there any problems?

Thank you very much!

Yikes. That means that you’ll be generating a matrix of 61,160 x 61,160 floats. That’ll definitely blow memory - with the default qiime2 workflow. I’d suggest trimming down that tree to something like 10,000 tips (i.e. by aggressively filtering out low abundance species).

That being said, we do have most of the machinery in place to do a sparse ilr transform (in fact the sparse ilr basis can be found here). So this will definitely be on the horizon!

1 Like

Hi Morton,
Thank you very much for your reply.
I’ve read the filter data part in tutorial, but it is all about filtering the table. Is there any methods to filter tree within the q2?

Thank you very much!

Not directly — filter sequences, then build a new tree.

I hope that helps!

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.