RNA-seq - Biological replicates not correlating! - Spearman Rank Correlation

Question

[Vignesh]

After performing pooled assembly using Trinity, I estimated the abundance for the transcripts using Kallisto for 4 samples. Each sample has three replicates so a total (4x3=12) samples. I checked the correlation between replicates using Spearman Rank Correlation method using the Estimated counts for all the samples. But the replicates are not correlating very well. ENCODE suggests that the correlation value should be from 0.92 to 0.98 or else the process should be repeated. Mine comes around 0.60 to 0.65, not even close. What might be the problem? Is this because of a sequencing error or any other problem? Can anyone kindly help?

Answer 1

[ATpoint]

Some details on the experiments would help to interpret what you say. What kind of replicates are this? Bological, technical (=sequencing replicates?) My personal opinion: Forget ENCODE guidelines and rank-based correlations for experiments with that many datapoints as RNA-seq. Better look at your data and then decide. How do they cluster in a PCA, do biological groups cluster in proximity? Is the mapping rate to exons (or in your case the % reads assigned to the transcriptome) good, do canonical genes that must be differentially-expressed based on experience/literature indeed come out as such?

[Kevin Blighe]

As per my colleague, do not just make a conclusion based on a correlation value. You should aim to explore multiple lines of evidence before making any final conclusion. This is part of research and part of your [I assume] learning experience working with data.

Also, for queries like this, you should really show exact commands that you have used.