add pyDeseq2 variant of analysis

[enryH]

• used Claude to translate Rmarkdown to Notebook
• made small adaptions, but the equality of the results has to be still tested

Concepts form R seem to be translated to Python.

• add Python to dockerimage with dependencies
• ensure equality of results

[enryH]

I left the output in the notebook so it can be easily inspected:

https://github.com/biosustain/dsp_transcriptomics_training/blob/add_pyDeseq2/01_scripts/02_differential_expression_analysis_pyDeseq2.ipynb

or using the nbviewer:
https://nbviewer.org/github/biosustain/dsp_transcriptomics_training/blob/add_pyDeseq2/01_scripts/02_differential_expression_analysis_pyDeseq2.ipynb

[Copilot]

Pull request overview

Adds a Python/pyDESeq2-based differential expression analysis workflow alongside the existing R DESeq2 report, aiming to reproduce the treatment vs control contrast in a Jupyter-friendly format.

Changes:

• Introduces a Jupytext-compatible Python notebook script implementing DE analysis with pyDESeq2 (including shrinkage, plots, and result export).
• Loads nf-core merged gene counts + sample metadata, constructs a pyDESeq2 dataset, and runs the treatment vs control contrast.
• Writes full and significant results tables and optionally compares overlap with existing R outputs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sebschulz1]

It might be useful for some users to add a link to the microforge site to get installation instructions immediately (if they don't have conda installed yet)

[sebschulz1]

Testing procedure for comparison of pydeseq2 and R-DESeq2 results
I have pushed an analysis script and results to this PR. Results from the R and the Jupyter notebooks (both deposited in this branch) were compared by calculating difference and ratio of log2FC and padj values. Additionally, scatter plots were generated to compare original log2FC and padj values visually.
I compared the results of all genes and the ones from only significant genes.

Main findings
In general, the agreement between both packages seems to be good. However, additional in-depth analysis could be performed.
I noted that certain genes are in output tables from pyDESeq2 but not in those from R-DESeq2 (missing values for such genes were replaced by zeros to be able to plot them). This observation was discussed with @Juassis and it likely stems from a pre-filtering step in the R scripts to delete low-count reads. However, this needs to be verified.