Interesting Genes

[orenbenkiki]

Compute the how much each gene is "interesting". To keep the cost of computation down, start with only genes which have sufficient expression (consider using some of the mean, variance, top-3, etc.). For this short list, compute how "interesting" the gene is using one of the following methods (needs investigation - the cheapest one to compute the works well?)

• Divide the UMAP projection into N*N squares, compute mean expression in each square, compute entropy of this for all squares that contain any metacells
• Use the skeleton KNN graph, for each metacell compute its connected (probably connected-to-connected) set of metacells, compute mean of the gene for each such set, compute entropy of the means across all sets
• Compute 2D concentration of the gene expression level in the UMAP (Akhiad's suggestion - seems there are built-in functions in R to do this which aren't too expensive).
• Some other method which is cheaper and gives a similar result?
  Either way, use this as the default order of genes shown to the user so that "interesting" genes would be easy for him to find.