Read count difference between STAR and Bowtie

Question

[hothriananya]

Hi, I aligned a small RNA seq data with Bowtie and STAR and I observed that there is a large difference between the read counts from both the aligners for the same miRNA from the same sample. I further checked the bam files of the same sample from both the aligners and found many differences. I am trying to understand what are the major reasons for these differences. Is it majorly algorithm differences? What are the other criteria I can consider to understand the read count differences? I am posting a picture of igv viewer of bam files from both the aligners for reference

Thank you, Hothri

[WouterDeCoster]

How to add images to a Biostars post

a large difference

found many differences

Please be more specific.

[Martombo]

why don't you use a program designed to align and count miRNA reads like miRdeep2 for example?

[hothriananya]

miRdeep2 is very specific to miRNA only, I am looking for small RNA expression ( rRNA, snoRNa etc)

[GenoMax]

Please add exact command lines used for both programs.

[hothriananya]

bowtie ${GENOME}/GRCh38.91 --threads 16 -S -q -n 2 -l 28 -e 70 -S ${FASTQ}.sam

STAR  --genomeDir ${GENOME} --readFilesIn ${FASTQ1} ${FASTQ2} --readFilesCommand zcat --runThreadN 8 >> STAR_progress.txt

Answer 1

[jared.andrews07]

Bowtie by itself should not be used for RNA-seq alignment, as it doesn't handle splice junctions and will fail to map a fairly large number of reads that overlap exons. Tophat uses bowtie for alignment, but uses additional transcriptome information to align reads near splice junctions more properly. However, newer aligners like STAR or HiSAT2 are much faster and generally better aligners that do a much better job of mapping to splice junctions than Tophat/bowtie. They are also much better at mapping lower quality data. Stick with STAR, though your count differences in your example there aren't actually all that extreme.

[hothriananya]

This is small RNAseq data, I used bowtie to align short reads.

[ATpoint]

As Wouter said, please specify what the evidence for the differences are beyond manual inspection on the browser. Given the screenshot, I see a read count difference in that miRNA of 6 out of 5600 which is 0.1% and therefore pretty much nothing. Please elaborate.

[hothriananya]

I found huge differences in counts from both the aligner, Below are the read counts for few miRs. Read counts from STAR aligner are higher than the Bowtie aligner read counts. These miRs are expressed in one of the control samples.

    STAR    Bowtie
MIR130A 117.49  73
MIR103B1    5776.49 303.773
MIR106A 14.31   6
MIR151A 878.233 369
MIR30C2 1820.4  104

[jared.andrews07]

Like I said, the aligners are quite different, so there will be some differences, particularly for something so small as miRNAs. Using a miRNA-seq specific aligner as suggested by Martombo in his comment may be your best path moving forward.

[GenoMax]

It is fine to use bowtie v.1 for miRNA alignment since those need ungapped alignments.

[jared.andrews07]

Yes, I didn't realize it was miRNA-seq originally. Good to know!

[WouterDeCoster]

I added markup to your post for increased readability. You can do this by selecting the text and clicking the 101010 button. When you compose or edit a post that button is in your toolbar, see image below: