Dear all,
I am quite new in bioinformatics/data science and I need some help with this issue. While I understand the intuition behind shrinking coefficient estimates and why this is applied to RNA sequencing experiments (I've studied Lasso and regularisation before), I don't understand the Bayesian approach to shrink the dispertion and LFC parameters. A good tutorial or a brief explanation here will do.
Thank you!
I have read the paper, it's self-explanatory! The only part I don't understand is how Bayesian shrinkage is performed. I have been following your bioinformatics repository on Github, it has been really useful. What I understand from Bayesian shrinkage is that we start from a distribution of MLE LFC estimates. We then fit a normal distribution of mean 0 to these estimates that we use as prior, right? What I don't understand is how the posterior distribution is calculated and how the normal distribution with mean 0 helps shrink the LFC estimate.
Those details are in the two methods papers (note we use a Cauchy now, not Normal... see paper 2).
The posterior calculation is all spelled out for Normal in the DESeq2 paper.
In apeglm we approximate the posterior using the Hessian at the MAP.
For motivation on why do hierarchical modeling, there is a nice paper from Hongkai Ji and Shirley Liu. Just google their names + hierarchical model.
Thanks, I will have a look. Do you recommend other resources besides your GitHub repository to get me trained up as quickly as possible?
Thanks!
Dr. Ramon Soto
Sorry I’m pretty busy and can’t reply more here. Ive listed plenty of references.