WIP: ILR transform on tensors #75
Conversation
mortonjt
requested review from
thermokarst and
ebolyen
and removed request for
thermokarst
|
Hey @mortonjt - the PR submission cutoff for the 2021.4 release was last week - I don't think we'll be able to squeeze this into this release. The next release is slated for August. |
|
@thermokarst the August release timeline sounds perfect :) |
| trimmed_tree = rename_clades(trimmed_tree) | ||
| # TODO: watch out for this indexing with diff. The ordering of the dims | ||
| # could mess things up here. | ||
| posterior_data = np.array(dataset['diff'].values) |
There was a problem hiding this comment.
we should replace this with a pandas pivot so that the dimension order doesn't matter.
|
Just raised a PR here on the tensor types: qiime2/q2-types#261 |
|
@thermokarst just a heads up on this. After talking to @ebolyen and @nbokulich, I think the most reasonable approach is to address qiime2/q2-types#261 by first creating a separate q2-bayes-types repository with the appropriate tensor types in place. That repository still needs to be fleshed out, so I don't think that can be done before the next release. |
|
Hey @mortonjt - we are doing some PR triage and review right now, and I looked over this PR along with the one from q2-types you linked above (#261). Are you still building that custom repo for the types you discussed with @nbokulich? And if so, are the changes in this PR still current? |
|
Hi @lizgehret thank you for following up on this. I had some q2type work on mortonjt/q2-types@060acfe But the effort was proving to be too much for me to work on at this point -- I would like to revisit it at a later date; but it will require me to either reprioritize projects or have someone else lead. |
|
No worries @mortonjt! I'm going to close this, but feel free to re-open if the situation changes. |
This addresses #74
The idea here we want to be able to apply the ILR transform directly to the outputs of Bayesian differential abundance tools such as Birdman. Doing so will allow for robust hypothesis testing, uncertainty estimation of balances in addition to providing an additional link between differential abundance tools and Empress.
Of course, this all depends on finalizing the upcoming tensor types https://github.com/gibsramen/BIRDMAn/issues/26
In the most ideal scenario, this would be included in the August qiime2 release, since we are aiming to include this functionality in upcoming manuscripts / plugins.