Usage with small number of cells

[tjbencomo]

Hi. I'm trying to use PISCES to analyze 10X single cell data from 5 tumor samples. Some samples contain significantly more cells than others after quality control filtering. The number of cells per sample ranges from approximately 600 to 7300 cells. Based on gene expression clustering, cell types are also disproportionately abundant in the samples. One sample only has 150 epithelial cells while others have upwards of 1500 epithelial cells.

In your recent paper using the PISCES pipeline to identify recurrence-associated renal tumor macrophages, I noticed you generated ARACNe networks for each sample separately. I am having trouble following the PISCES guideline to generate metaCells from clusters with at least 500 cells because the primary population of interest (ie epithelial cells) doesn't have enough cells in some samples.

How would you approach this problem? I had a few thoughts:

1. Pool samples together and create one set of ARACNe networks for all 5 samples rather than patient specific networks
2. Use all samples and only analyze cell types with enough cells to generate patient specific networks (ie don't analyze epithelial cells and only focus on immune components)
3. Discard samples with too few cells and only analyze samples with enough cells for all cell types of interest