Entering edit mode
3.0 years ago
lucia.liu54
•
0
I am analyzing differential expression among three different factor levels (control samples, samples with overexpression (OE) of gene of interest, and samples with knockout (KD) of gene of interest). I understand I can use the likelihood ratio test to look for general expression changes among the three conditions at once. However, could I also intersect pair-wise Wald test results for control vs OE and control vs KD? E.g., genes that are both significantly downregulated in knockout samples and upregulated in overexpressing samples (compared to control samples).
Yes, there is no problem with taking an intersection of results. The LRT does not test against a fold change but only for the goodness-of-fit, so I personally find the Wald test more intuitive to interpret, especially with a limited number of experimental groups. Just try it and see whether it answers your scientific question. It always (for me) makes sense to take the union of all DE genes of all groups and make a hclust-clustered heatmap to visualize the general behaviour across the samples, so seeing trends between groups.