human RNA seq alignment QC

[liumangmang]

Hi,
I used HISAT2 to align human RNA-seq data (Illumina PE, Stranded, rRNA depleted).
We are interested in comparing gene expression levels, snp and different isoform usage between treatment conditions in the end.
I got about 70-75% reads aligned concordantly 1 time and about 20% concordantly >1 time with less than 10% disconcordantly aligned. Overall mapped rate is over 95%.
Is there a common benchmark that how much % unique concordantly mapped reads is typical for human RNA?
I'm wondering how to determine whether alignment result is acceptable for carrying on the downstream analysis and how I could tweak with the parameters in HISAT if needed.....

Thank you,

[SylvainL]

Sounds good to me... Over 70% of uniquely mapped reads for Human is quite good. You also have to keep in mind that this percentage can obviously depend on the expressed genes...

[Michael.Ante]

Hi,

Before alignment I'd perform QC with FastQC and FastQC screen (especially in case of rRNA depletion).
After alignment, I'd perform QC with a set of RSeQC's tools:

• bam_stat.py
• clipping_profile.py
• geneBody_coverage.py
• infer_experiment.py (if your library was strand-preserving)
• inner_distance.py (for PE-runs)
• read_distribution.py

A lot of results can be collected by the multiQC reporting tool.

Cheers,
Michael

[Persistent LABS]

Hi liumangmang,
Your alignment summary looks good. There is a similar post here: http://seqanswers.com/forums/showthread.php?t=29769
You can refer this.

[liumangmang]

Thank you all, I'll try those QC tools and come back to update.

[liumangmang]

Update:
We get average about 15%-25% multi mapped reads. And there are some public available data from SRA (same tissue type and protocal), also get similar multi mapped ratio.
And RSeQC + multiQC really worked well, it gave me pretty summary report!

Thank you all.