Mapping to transcriptome with bowtie2 vs mapping to genome using STAR

Question

[GouthamAtla]

Hi,

Does any one know any paper or personally compared the results of the following approaches ?

1. Mapping RNA-Seq reads to transcriptome fasta using bowtie2 and quantifying using RSEM or other tools.

2. Mapping RNA-Seq data to genome fasta using STAR and quantifying using featureCounts.

In first case, there is less chances of finding multi-mapped reads because we focus only on transcribed regions and ignore rest of the genome. Don't know how often it's a problem or meaningful.

Some of the data I am looking at shows differences in quantifications coming from two methods for some important genes, so just thought if some one has more insights into it.

[Devon Ryan]

Have you looked at one or two of those genes in IGV? Are the reads to them multimappers (presumably due to a similar pseudogene)?

[GouthamAtla]

I STAR re-aligned using very low stringent options, no luck so far. I am going to try exact method the authors described in their paper.

Answer 1

[GouthamAtla]

figured it out. Sometimes annotations includes 3' UTRs. So if the annotations we are using includes 3' UTRs, we get counts for genes. If not, we dont quantify them. Its not problem of mapping but the annotations we are using.

Here Refseq annotation did not include 3'UTR but the Gencode. So the quantifications differ based on this.