Dealing with missing values in differential uncertainty test

[emdann]

i.e. when a sample has no cells in a nhood? Possible solutions

• Set confidence to 0: suboptimal, leads to false negatives when all the other values for the condition are high and values for control are lower than existing values but higher than 0
• Set confidence to mean of other replicates: this would make sure the missing values don't affect the mean value, but it would affect the variance
• Exclude neighbourhoods with missing values by filtering / using bigger neighbourhoods: somewhat suboptimal, esp if only one rep is missing you reduce the space where you can do test
• Use masked array or exclude missing samples somehow: means you have different power in different regions of the graph (but this is also the case with different no of cells, so maybe not a massive deal)

[emdann]

Use masked array or exclude missing samples somehow: means you have different power in different regions of the graph (but this is also the case with different no of cells, so maybe not a massive deal)

Going with this option for now, with https://github.com/emdann/diff2atlas/commit/c9a9a9f4932903104847b88a1870a7857063d5fb