I am trying to do DE analysis of non-coding features of A. thaliana. I find in the miRNA and lncRNA counts file that they are abundant in zero counts, and most of the non-zero counts are very low. Now, for the DE analysis of coding features, an inbuilt filtering method to remove the extremely low counts is available in edgeR, called filterByExpr(). Is this function suitable for only those features with high counts like coding sequences, or is it suitable for non-coding features with low counts and low library sizes too? If not, then what kind of filtering criteria should I use for filtering out genes with insignificant counts? Or should I filter out only the ones with row sums 0?


1 Answer 1


Partially answered before. I am assuming this is a lncDIFF calculation. For the lncDIFF paper it is unlikely to work as the following figure displays,


However, don't take my word for this, test it. You should test the option explicitly via glmQLFTest and glmLRT. If you perform either of these tests with or without the filter the better fit (one not rejected) is the correct answer. I strongly predict the filtered data will be rejected and unfiltered data (hopefully) will pass.

The question of what is better glmQLFTest versus glmLRT is discussed in the previous answer, here Statistical methods suitable for DE analysis of non coding RNAs

Just keep in mind this is plant data and these tests are developed for humans.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.