Hi Everyone,

I'm trying to analyze some RNAseq results, but one of my samples is a pretty bad outlier by PCA and by clustering over the entire transcriptome.

I have 4 groups with 3 biological replicates.  These samples were run in 2 batches.

When I try to summarize my reads using RSubread's featureCounts, the outlier has a very low assignment %, with a high % of multiple assignments

My question is how should I proceed with my analysis?  I don't have enough replicates to kick the outlier out.  Are there methods to fix outliers?  Is it valid to consider this outlier as a separate batch (Removing the variation with removeBatchEffect)? 

Thank you!

voomWithQualityWeights to the rescue!

... in the limma package (in case you weren't aware).

No, it isn't valid to consider an outlier as a separate batch.

There are only two possibilities: down-weight the outlier using the appropriate functions in limma (as suggested by Steve) or throw the sample out (as suggested by Dario).

Since you have three replicates in each group, you should just exclude the unusual sample from the analysis. You don't need balanced sample sizes in every experimental group.