Logit-normal distribution to model variation among biological replicates

Question

[CY]

I came across rMATS while doing research on differential alternative splicing (AS). It uses logit-normal distibution to model variation among biological replicates. I am having difficulty understanding the logics here.

Can anyone share some insights on how logit-normal distribution can model such variation? It does not have to be AS specific as long as the allow me to understanding the rational here.

Answer 1

[Kevin Blighe]

It says, in the paper:

rMATS uses a binomial distribution to model the read count from the exon inclusion isoform given the exon inclusion level in each individual replicate and a logit-normal distribution to model the variation among replicates within sample group.

This is a big assumption that likely does not hold for all genes, which is a limitation of rMATs. I guess that they mean, that, if you plot the read counts for every exon over every identified isoform, then it would follow a binomial. Exons that are only expressed in rare isoforms would be at either tail of the curve.

[CY]

Thank you for replying. I can understand the binomial distribution part. What confused me is the logic of using logit-normal distribution to model between-replicates variation