how to evaluate if the replicates are good

[capricy]

We have 5 replicates for each sample. Before we go with further analysis, we would like to know if those 5 replicates agree with each other in a reasonable way. So is Pearson correlation suitable in this situation?

Thanks for any input

[gringer]

is Pearson correlation suitable in this situation?

My guess would be no, even if you're using log-normalised expression values (I'm assuming here this is an RNASeq project). It would be expected, even in substantially different samples, that the majority of genes would be similarly expressed, so you'd always get your samples appearing to be in almost complete agreement with each other (with crazy-low associated p-value).

[capricy]

Then what should I do? Could you please give me more details? Thanks.

[gringer]

Then what should I do? Could you please give me more details? Thanks.

It seems like you're asking me specifically. I have limited experience in this area, so can't provide useful information to you on this; there are much more qualified statisticians on this forum who can give better advice.

The only thing I can quickly think of is doing a differential expression experiment with no replicates, to see if any genes come up as differentially expressed between any replicate pairs. Unfortunately, that will tell you if your samples are different (which you really hope is not the case), but not if they are the same.

[dpryan]

If you just want to see if the samples within each group have reasonable similarity, then you might just use hierarchical clustering or look at a PCA plot. Those are normal QC steps and there are probably example steps for whichever analysis program you use (I recall there being examples in the edgeR and DESeq vignettes).