v0.0.10

[cameronmartino]

v0.0.10 (2023-12-01)

Bug fixes

• Removed np.int types
  • fixes bug that prevents use in newer versions of numpy / QIIME2
    • see issue possible numpy-related issue #71

Features [experimental]

• rpca projection of new data and with cross validation
  • A new function for RPCA where new data can be projected into an existing ordination.
  • Also allows for internal CV, where the hold out data is projected into the training ordination, which is then used to re-build the test data, and the error between the projection and the real test data is calculated.
    • see issue low-rank determination not working with high-rank data #70
• tempted.py and assocated tests/commands/tutorials
  • Added in a python implementation TEMPTED, more details on the methods inm this paper here.
• joint-rpca and associated tests/commands/tutorials
  • Joint-RPCA is an extention of RPCA for multiple omics types.
• qc_rarefaction
  • This function compares the mantel correlation of distances to the abs. differences in sample sum between rarefied and unrarefied input data. This is an easy check to ensure the results are not bieng significantly altered by non-rarefaction is cases of large differences (e.g., low-biomass) or where the sample sum differences can not be seperated from the phenotypes/low-rank structure of the data (i.e., deep sequencing of controls and shallow of sick).
    • see issue low-rank determination not working with high-rank data #70

Deprecated functionality

• auto_rpca across the package and rank_estimate in optspace.py
  • was not performing correctly and no reasonable quick fix is available.
    • see issue low-rank determination not working with high-rank data #70

[SilasK]

Looking forward to the new version.

[SilasK]

I'm testing this branch and have some comments.

Don't forget to include in the release notes the breaking change from

from auto_rpca to gemelli.rpca import rpca

Using simulated random data I get the

gemelli/gemelli/preprocessing.py:426: RuntimeWarning: divide by zero encountered in log
  mat = np.log(matrix_closure(mat))

[cameronmartino]

Thanks, @SilasK! That warning can be ignored, it is meant to be suppressed within Gemelli but if numpy is re-imported before running RPCA it will get un-suppressed. Not sure how to fix that or if I want to since the user might want that warning for other parts of their own code. It comes from the RCLR command where the infs/-infs/NaNs are later ignored. If you want to suppress it in your scripts add

numpy.seterr(divide = 'ignore')

after your numpy import.

Let me know if you run into any other issues/comments on the new branch!

[SilasK]

Maby there is a divide function that can handle nans.

[SilasK]

Otherwise I could use it.

[gibsramen]

Self-assigning for review

[gibsramen]

I have no idea why, but on my end the Joint-RPCA CLI notebook fails at the PERMANOVA step.

I managed to fix this with the following change but it's unclear why it even failed.

Pandas version is 1.5.3

[cameronmartino]

@SilasK looking into it seems like there are some issues around that (here). That may be why we opted for the way we do it, I don't remember. I opened an issue for it and will look more deeply into this for a future version. Thanks for pointing it out!

[cameronmartino]

@gibsramen Fixed, thanks!

[gibsramen]

Couple changes requested, thanks!