Sample-dominated neighborhoods

[gcgreenberg96]

Hello,
Are there any recommendations on how to deal with sample-dominated neighborhoods? Even when using a very high n_neighbors, there are many. In my analysis, most regions that appear to be DA turn out to be highly skewed by 1 or 2 samples. The regions with low LISI scores (indicating sample-dominated) highly align with where neighborhoods are DA (as seen below).

Some idea:

• Remove all sample-dominated neighborhoods
• Remove all cells with low LISI (or other metric)
• Make the linear model more robust to outliers (many options for this).

Thanks!

[MikeDMorgan]

Hi @gcgreenberg96 This is usually an issue of data integration and/or massive imbalance in cell number between samples rather than an issue with Milo or DA testing. You could test the latter by randomly downsampling the number of cells in each experimental sample and repeating the integration, DA testing, etc.