Discrepancy in Summation Results When Fetching Matrix Data with Different Methods

[skelviper]

Discrepancy in Summation Results When Fetching Matrix Data with Different Methods

[nvictus]

Actually, there is no discrepancy.

The "as-pixels" selector clr.matrix(balance=False, as_pixels=True) only returns the upper triangle portion of the global contact matrix, including the main diagonal, which will correspond to what you see in a pairs file.

clr.matrix(balance=False) returns a numpy array of the full contact matrix, including the lower triangle portion as well. At lower resolutions, more counts get absorbed into the main diagonal, which naturally changes the way these add up.

You can validate this:

for res in cooler.fileops.list_coolers("test.mcool"):
    clr = cooler.Cooler(f"test.mcool::{res}")   
        
    df = clr.matrix(balance=False, as_pixels=True).fetch("chrX(mat)")
    diag_sum = df[df['bin1_id'] == df['bin2_id']]['count'].sum()
    upper_sum = df[df['bin1_id'] != df['bin2_id']]['count'].sum()
    array_sum = clr.matrix(balance=False).fetch("chrX(mat)").sum()
        
    assert(array_sum == diag_sum + 2*upper_sum)

Or by inspecting or visualizing the two results.

[skelviper]

Oh, I forgot about the issue with the upper triangle, that was really silly of me. Initially, when I saw that the matrix sum was different under different resolutions, I thought there would be some problems. Thank you so much for your help!