Question: What is the Median Normalized Average statistic in Loupe Browser?
Answer: The median-normalization average (MNA) of a gene in a cluster is calculated as the mean of observed UMI counts normalized by the size factor for each cell in the representative cluster. It is calculated as follows:
MNA(gene_X) = mean ([gene_X_UMI_count * size_factor for cell in cluster])
where, the size factor is computed for every cell as follows:
size_factor = (total_UMI_counts_in_current_cell / median_UMI_counts_per_cell)
For example, if a library has 1000 cells, then 1000 different size factors are calculated.
This calculation becomes apparent in Loupe Browser, when a user chooses between locally distinguishing and globally distinguishing categories of differential gene expression. For globally distinguishing, median_UMI count is calculated over all cells in the library. For locally distinguishing, it is calculated over the selected cluster subset. Thus the number of cells being considered changes in the size factor estimates.
References:
Github: https://github.com/10XGenomics/cellranger/blob/master/lib/python/cellranger/analysis/diffexp.py
Related kb article: https://kb.10xgenomics.com/hc/en-us/articles/360007388751-How-is-Log2-fold-change-calculated-