MSIsensor using variable numbers of positions

[RichardCorbett]

Hi there.
I made a number of random samplings of the full microsatellite list created with "scan". I am downsampling it to test if using a subset of the positions is still able to differentiate between high and low MSI samples. I'm seeing that even using a small number (10 thousand) positions I can easily differentiate the two groups of samples.

I have discovered something unusual though. For some reason, the high MSI cases get lower % MSI when I used more than 5000000 positions, and much higher when I use fewer positions.

Do you have any ideas of why higher numbers of positions for high MSI cases would not be in line with the lower counts of positions?