Normal topic model : voom variances or weights consideration

stephenslab / count-clustering

Code and data reproducing results from our PLOS Genetics paper "Visualizing the structure of RNA-seq expression data using grade of membership models"

http://stephenslab.github.io/count-clustering

5 stars 1 forks source link

Normal topic model : voom variances or weights consideration #3

Open kkdey opened 9 years ago

kkdey commented 9 years ago

The Normal Topic model has been found to do not so good a job at clustering samples as the Poisson model does. However it may be used for clustering CPM data or for batch corrected data. The question is how to improve the performance. One suggestion was to use the voom weights somehow. We need to figure out the model though.

jhsiao999 commented 9 years ago

Yes. The fact that the Normal Topic model doesn't do as well as the Poisson model may imply that we need to account for over-dispersion in sequencing data. What about the existing methods? Do they account for over-dispersion? (clustering methods only)