I am testing the various featurecounts options from the very useful Rsubread package.

I have noticed that in all tested libraries (>100) using exon + useMetaFeatures (Ensemble gtf, GTF.featureType = "gene", GTF.attrType = "gene_id") assigns almost twice as many reads to features (metaFeatures) compared to exon without useMetaFeatures (all other settings are identical).

I know that the intent of feature and metaFeature read assignment is different but this vast difference puzzles me since the metafeatures are produced by combining the same exons.

I expected a reduction in assigned reads but the difference is quite large.

Anyone?