Done on the Tabula Muris Senis dataset.
Data is raw RNA-seq counts from

https://czb-tabula-muris-senis.s3-us-west-2.amazonaws.com/Data-objects/tabula-muris-senis-facs-official-raw-obj.h5ad

Visualization of the result can be found at https://gos.garden/vis-age/
The visualization code - in https://github.com/justgos/vis-age

Pre-processed as

• RNA counts normalized to 1.0 per cell

Cell statistics (expression mean and std.dev) are calculated - with 4 levels of granularity - on subsets defined by a unique combination of

• Animal sex
• Tissue type
• Subtissue type
• Cell type

Expression fold-changes computed between subsets of the pairs of age groups are filtered as

• Subset has >= 3 cells
• Expression is >= 1e-12 in both age groups
• Bonferroni correction is applied to p-values by multiplying them first by the number of genes (~20k) and then by the number of subsets (2-1000 depending on granularity). Results are then thresholded by 1e-5

Notebooks

• cognitis.ipynb - exploration + systematic analysis of expression fold-changes
• videns.ipynb - preparation of additional pathway data for visualization

All the results for different analysis granularity levels:

• results/merged-cell_ontology_class.csv
• results/merged-subtissue.csv
• results/merged-tissue.csv
• results/merged-sex.csv
• results/merged.csv - total merge across all the granularities