Discordant results from the Scanpy tutorial "Preprocessing and clustering 3k PBMCs"

Question

[sbstevenlee]

Hello,

I'm new to single-cell RNAseq analysis and have been trying to learn how to use the scanpy program. To this end, I wanted to start by following the Preprocessing and clustering 3k PBMCs tutorial. So I cloned the entire GitHub repository ("scanpy-tutorials") to obtain the Jupyter notebook ("pbmc3k.ipynb") and then ran it without any issues.

Most of the results look consistent with what's shown in the tutorial page, but I noticed that my UMAP visualizations look dramatically different from that of the authors. See below comparison.

From the website:

From my notebook:

Can anyone please tell me why I'm seeing this difference?

1. As I mentioned above, I cloned the entire repository and didn't make any changes to the notebook.
2. I'm using the latest version of scanpy (1.7.2).
3. I noticed that the tutorial seems to be running 1.6.0, but when I installed that version and reran the notebook, it gave an error upstream.

Answer 1

[sbstevenlee]

Answering my own question:

A colleague of mine told me that a newer version of scanpy is available in the conda-forge channel (1.8.2), so when I downloaded it and reran the notebook it produced plots that are much similar to the original ones as shown below.

I'm still not sure why the 1.7.2 version produced the awful-looking plots, but at least the problem is gone. I also want to note as a record that I downloaded the 1.7.2 version from the bioconda channel (currently, it's the most latest version).

[rpolicastro]

bioconda recommends having the conda-forge channel as a higher priority when installing any software btw.