Library Depth Normalization

[imk1]

I noticed that you have multiple options for library depth normalization, including upper quartile, total sum, and harmonic mean, and that upper quartile is the default. Can you explain my upper quartile is the default? (My intuition is that total sum better captures the library depth.) Thanks!

[talashuach]

Library depth effects can be seen as shifting the entire distribution of reads-per-barcode. So in the general case, an upper quartile and the total sum should return very similar correction factors.
However, if some samples have extreme outliers (which I've seen in some MPRA datasets), the total sum can be pretty sensitive to those, whereas the upper quartile is not. So by default we use an upper quartile, but total sum is supported if you know your data doesn't have this issue and prefer to use the entire sum.

Hope this helps!

[imk1]

Thank you for this very helpful explanation! Upper quartile seems like it would handle extreme outliers on the low end of reads but not on the high end of reads. Is there a reason for not also eliminating the top few percent?