MethylVI batch effect tutorial

[ethanweinberger]

Tutorial accompanies scverse/scvi-tools#2834

Formatting

• My tutorial has only one top-level (#) header

Reproducibility

• My tutorial works on Google Colab
• My tutorial sets scvi.settings.seed = 0 at the beginning of the notebook
• My tutorial has been run and includes outputs (e.g. plots, tables)

Other

• Counts and normalized data should co-exist in the datasets, see the API overview for an example
• For scRNA-seq data, normalization should be counts per median library size and then log1p transformed -- if not, a reason should be given (Note: not applicable to this tutorial as we're not using scRNA-seq)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[canergen]

@ori-kron-wis can we use our download function instead?

[canergen]

jointly = one model but treat both seperately not add both and get one output? Maybe write "we'll jointly model both".

How would modalities={"categorical_covariate_keys": "mCG"}look like for two different fields. Please put in the comment.

Reply via ReviewNB

[ethanweinberger]

To make the description clearer, I changed the wording to "Here we'll jointly model both CpG and non-CpG methylation features" as suggested.

For the modalities argument, I added a {note} block similar to the one in the totalVI tutorial (https://docs.scvi-tools.org/en/stable/tutorials/notebooks/multimodal/totalVI.html)

[canergen]

Please add the code that you would use if mCH and mCH in the note block below. It doesn't become clear to me.

[canergen]

Maybe you post here the code if both have different categories and then I can advise what's the minimum information required.

[canergen]

Maybe show get_normalized_methylation or DE additionally to go beyond latent space analysis? See our other tutorials for examples.

Reply via ReviewNB

[ethanweinberger]

Once this notebook is merged in I plan to create a separate tutorial for feature-level analysis that will replicate the results from our (upcoming) preprint

[canergen]

Can we have it in one notebook? I don't like bloating the number of tutorials. It's currently already quite painful to rerun and manually merge notebooks. Do you need different datasets for it?

[canergen]

See comments overall looks fine. @ori-kron-wis can you help make the style consistent with other tutorials (I assume removing notebook style commands). These might change rendering on the webpage, which can end up ugly.