I have duplicate samples for RNAseq and just wonder if there is way to test if the replicates are good or not and what criterion to use.

I remember there was a discussion about this recently in this email list, just couldn't find it. Wonder if somebody could give me a hint or point me to the old post.

Thanks a lot.

Standard exploratory plots will show whether the replicate libraries are reasonably consistent. For example (in edgeR or limma) MDS plots for MA plots would be used to explore this.

As Gordon says, MDS plots are great for examining replicate relationships. Other simple things to do would be to examine correlation among your replicates/samples. For instance, given a table of counts:

heatmap(cor(counts), scale="none")

Or a heat map with a legend:

library(gplots)
heatmap.2(cor(counts), trace="none")

Pairs plots also come in handy to examine differences between replicates vs. conditions. For instance if you had two replicates of two conditions:

pairs(log2(counts+1))

would generate pairwise scatter plots of all columns (adding 1 avoid the problem of taking the log for genes with 0 counts). Hopefully your replicates would not show more scatter when compared to each other versus compared to a different condition.