As an important aside for any differential co-expression analysis: while we often hope that the latent factor driving differential co-expression is biological, it is very difficult to rule out a latent technical factor.

from shared unmodeled biological variation involving an observed risk factor and a latent factor. Our approach, KDCA, uses a flexible kernel-based framework to detect complex differential co-expression patterns while handling continuous or categorical risk factors and batch effects!

Software: github.com/ajbass/sffdr

We have software you can use on GitHub: github.com/ajbass/kdca. Work with @epsteinstatgen.bsky.social and @davecutler.bsky.social.