search

davidaknowles / leafcutter

Annotation-free quantification of RNA splicing. Yang I. Li, David A. Knowles, Jack Humphrey, Alvaro N. Barbeira, Scott P. Dickinson, Hae Kyung Im, Jonathan K. Pritchard
http://davidaknowles.github.io/leafcutter/
Apache License 2.0
207 stars 115 forks source link

modeling random effects #44

Open ddpinto opened 6 years ago

ddpinto commented 6 years ago

Hello David, Is there a way to model individual as random effect as part of the differential splicing analysis, so that one could account for non-independent samples (duplicates) or repeated measurements for the same individual?

davidaknowles commented 6 years ago

Unfortunately it's not straightforward to include random effects in the Dirichlet-Multinomial GLM. The downside of the non-Gaussian likelihood is that random effects cannot be analytically integrated out (like you would for an LMM for example). There are approaches to perform the integration, using quadrature/Laplace approximation/variational methods/MC sampling but all of these trade-off computation time and accuracy/robustness, so it would require some investigation to work out the best way to do this.

I'll leave this open for now and maybe people can thumbs-up your comment if they would also like this functionality.