confusing results from EBSeq-HMM (down down vs up down)

[Christopher Balakrishnan]

Hi there, we've been trying to use EBSeq-Hmm to look at data from a 3 point time course of sorts. n=3 per time point.

We are getting results that make sense biologically (GO categories and such), but the categorization seems odd. For example, the attached plot is classified as a "down-down" gene, whereas it would appear to me to be an up-down gene.

It is possible we are doing something silly, but if anyone has an obvious idea what that might be it would be helpful to trouble shoot. Or is the sample size just too small? Has anyone else had similar issues?

Thanks!

Chris

[Christopher Balakrishnan]

Update: using more stringent FDR cutoff seems to help. My impression though is that the down-down category can be strongly influenced by outliers (in our small sample size experiment). We get many more down-down transcripts than any other category, and most seem to have a high outlier at t=0. any suggestions to avoid this? (besides dong bigger experiments :) )

Chris

[Nereful]

Hi Christopher,

Were you able to address this issue?

I have similar concern. There is a contrasting results obtained between the estimated "Most Likely Path" and the projected trajectory (as visualized in a graph). Example, the attached figure has Down-down-down profile, but the visualized graph is not.

I tried to adjust the FDR value to 0.01 but I am arriving at a similar issue.

Thanks,

Nelzo