We performed RNA-seq data meta-analysis by the Rankprod (combined p-value) method to find DEGs. Surprisingly, we found some DEGs in meta-genes that did not have significant differential expression in separate study analysis (individual t-test).

Not disucssing that standard t-tests and RNA-seq are not a good combination, do you understand what a meta-analysis does and what combining p-values means? Say you have three studies and geneA had a p-value of 0.051 in each of the three. Not significant in each, but what would be the combined p-value, higher or lower? Think about that and then try to answer your question.

Thanks for your reply.
I do not know much about statistics. I just use the Rankprod method to do a meta-analysis. I used the combined p-value method. That was my question: why were some meta-genes left out of individual study analyses?

We conducted gene expression meta-analysis with Ranlprod method, too. and there are some genes in the meta-analysis output that was not significant in some of the individual studies. How could we explain it? what is the reason? That was expected variance between samples in each study was less than the variance between different studies and all meta-genes were common with DEGs in each study. how we can describe this result bae on variance change in the expression matrix?

I haven't done meta differential expression analysis before, but isn't the point of the approach to gain better sensitivity in DEG identification by using results from multiple studies? That's to say, it isn't that surprising that you found additional statistically significant results - even the authors of Rankprod highlight this in their paper.

From what I am seeing in the Rankprob paper the method does not combine p-values, but instead ranks genes by fold change values and performs a permutation test to determine statistical significance.

Not disucssing that standard t-tests and RNA-seq are not a good combination, do you understand what a meta-analysis does and what combining p-values means? Say you have three studies and geneA had a p-value of 0.051 in each of the three. Not significant in each, but what would be the combined p-value, higher or lower? Think about that and then try to answer your question.

Thanks for your reply. I do not know much about statistics. I just use the Rankprod method to do a meta-analysis. I used the combined p-value method. That was my question: why were some meta-genes left out of individual study analyses?

You just repeated your question without addressing my comment, I cannot help further it seems.

Hi

We conducted gene expression meta-analysis with Ranlprod method, too. and there are some genes in the meta-analysis output that was not significant in some of the individual studies. How could we explain it? what is the reason? That was expected variance between samples in each study was less than the variance between different studies and all meta-genes were common with DEGs in each study. how we can describe this result bae on variance change in the expression matrix?