LogFc vs t_statistic to estimate over/under expressed genes?

Question

[Manuel Sokolov Ravasqueira]

Hi!

I am doing DGE using limma+voom over two populations, I have a gene X that has significant adjusted p values yet logfc values of 0.6/0.7 which is below the threshold of 1.

Additionally I did t test over the Voom$E of the same two groups regarding the same gene x having the t stastitic > 0 and the p value significant.

Can I use this t test to determine over/under expressed genes instead of evaluating the p values alongside with logfc? I understand that I can just lower the threshold but I set the threshold according to distribution of logfc values and do not want to change it.

Best regards, Manuel

[bk11]

Please check in the link provided below why you can't use t-tests over the standard methods that already exists. We are blind about your sample sizes in the comparison groups, I suspecting a power issue in your experiment design which you can overcome by increasing the samples sizes. LogFC Threshold of 1 is not the universal rule for differential expression.

Why can't use t-test for differential expression (negative binomial distribution assumed)

Answer 1

[ATpoint]

How would that be any different from what limma gives you already (despite being less accurace)? There is no rule that 1 is the universal fold change. Overexpressed can simply mean logFC > 0 an underexpressed can mean logFC < 0. In general, when standardized tests exist, do not fiddle with custom solutions like t-tests. There is a reason why these specialized methods exist. The limma papers tell why that is.

[Manuel Sokolov Ravasqueira]

Thank you @atpoint. Then to analyze one gene in particular between two groups the best would be to plot a box plot of the distribution between two groups and have the p adjusted valued from limma and the logfc value?

[ATpoint]

Yes, that sounds intuitive.