Read counts generated from featureCounts

Question

[basuanubhav]

Hey all, I had a doubt regarding the output generated by featureCounts. So, what would be the difference in output between the following commands:

featureCounts -p -s 2 -F GTF -t exon -a annotation.gtf -o counts.txt sample.bam 

featureCounts -p -s 2 -F GTF -t gene -a annotation.gtf -o counts.txt sample.bam

From what I understand, the second command will only read the lines in the GTF which have 'gene' in the 3rd column while the first command will count the exon lines. Also, the sum of the reads for all exons of a gene should be the same as the number of reads for the gene as a whole. So, should the output file have the same counts for both?

Thanks,

Answer 1

[Carlo Yague]

From what I understand, the second command will only read the lines in the GTF which have 'gene' in the 3rd column while the first command will count the exon lines.

This is correct. Both will be aggregated on the gene_id metafeature (that is the default for option -g) in a 1 - 1 relationship with -t gene or in a n - 1 relationship with -t exon.

Also, the sum of the reads for all exons of a gene should be the same as the number of reads for the gene as a whole.

This is not necessarily true. -t gene will also count reads mapped on introns. Usually this will account for low read count because introns are often spliced out but there are many cases of intron retention, and in some species (such as mammals) average intron length is much longer than average exon length. Also because of splicing dynamic, it is possible to sequence unspliced or partially spliced pre-mRNA. So in general, you should not expect the option -t exon and -t gene to produce the same counts.

[basuanubhav]

Ah, thanks for the nice explanation. I hadn't thought of the aspect of reads aligning to introns skewing results. So, ideally, for an RNAseq experiment, I would want to use -t exon, correct?

[Carlo Yague]

Yes, definitely in most cases !

[basuanubhav]

Thanks a lot, cheers!

[Carlo Yague]

PS: unrelated to your question, but usually we do not use the -o option because it is often better to discard read that can not be unambiguously assigned, at least for differential expression analysis.

[basuanubhav]

Yes absolulety! Even I dont use the -O option. The -o I have used is just to define the output file.