comparison of BWA MEM and hisat2

Question

[guido.leoni]

I'm aligning mouse exomeseq (tumor and control) with bwa and hisat2. After comparison of the 2 alignments I see in many genomic positions strange reads with many mismatch aligned by BWA that are descarded by hisat2. Please could you give a feedback about the reasons for which BWA maps these reads? I suppose that them don't contain true variants but technical artifacts. The same reads are descarded by hisat2.

Here are the parameters for bwa mem and hisat 2

bwa mem: bwa mem -M -t 30 selection of reads samtools view -@ 30 -f 0x0002 -h -q 1 -F 4 -F 256

hisat2: --no-spliced-alignment --phred33 (remaining parameters default) selection of reads with NH:1/

I attach an example

thank you very much

[igor]

I see that the control has coverage of around 700 and the tumor of only 20 for both aligners. Assuming the overall coverage is even, it seems like both aligners are filtering out a lot of reads in the tumor. If that is the case, there may be something more complicated happening in that region.

[Chris Fields]

One thing that is missing is a 'normal' alignment using HISAT2.

Something that comes to mind is the low quality score threshold filter you are using. What is the MAPQ for those reads? Maybe a higher minimal MAPQ is needed, I've used MAPQ of 10 before.

NOTE: Heng Li had a number of great comments re: MAPQ, what an aligner calls a 'unique' alignment, and so on. If I can dig them up (or if someone else can find them) it would be great to post them. Would also be great to see how HISAT2 compares to BWA w/ simulated data and sensitivity/specificity.

[guido.leoni]

I added the control The MAPQ seems to me quite high (@ MAPQ 60)

[guido.leoni]

Thank you for your comment. The Control is an experiment taken from literature. It has very high coverage so I expected that the coverage of control was very high compared to my tumor