examination o gene expression by RNAseq: mapping to the genome or the geneset??

[capricy]

I used tophat to map my RNAseq reads.

1. mapped the reads to the reference genome and only examined the exon regions with the mapped reads. Then calculate the percentage of the genes covered by those reads (breadth coverage).

2. Extracted the exons to generate a reference geneset. Then mapped the reads to this geneset and calculate the percentage of the genes covered by the reads

The results from 1 and 2 were compared and they were quite different.

For my specific data set, 2 generally gave a higher coverage percentage when compared with 1.

Geneset accounts for 5% of the total genome.

I wonder if it is because genome is too big and the mapped reads are more spread along the genome especially for those duplicate/repetitive regions. If geneset is used as reference, reads are forced to mapped to the limited regions and the breadth coverage is increased accordingly???

Any thoughts about this??

[dpryan]

Yes, mapping to the apparently expressed transcriptome may increase the mapping rate, at the cost of increasing the false-positive alignment rate. This is sort of the opposite way in which tophat2 works.

[capricy]

But why do I think that mapping to sequences with apparently high background noise will increase the false negative?

[dpryan]

It'll likely do both. Aligners need to make a trade-off between speed and accuracy. If a seed matches more than, say, 20 times, then following up on all of the hits might take too long to be worthwhile (this will increase the non-alignment rate). Similarly, if there's a mismatch in the seed, the search-space may go up drastically. Those are among the reasons that aligning to the genome can also increase the false negative rate.