Forum:[Opinion] - What alignment software do you use?

Question

[Andrew_MacGregor]

I've read a number of papers that compare the dozens of aligners out there. In general, they seem to agree that there is no "one best aligner" and that different software packages have different niches.

It seems like the general consensus is that bowtie2 and BWA-MEM are both great all-around tools.

When setting up your data processing pipeline, did you pick a good all-around aligner like bowtie2? Or did you experiment with other aligners before settling on something that works the best for your needs on your hardware?

[WouterDeCoster]

Eh obviously the best aligner always depends on the application, for example for long reads I'd say minimap2, for short spliced alignment STAR.

[GenoMax]

If you are only interested in specific edge cases then you may need to evaluate many. Otherwise most aligners should generate comparable alignments (within constraints of selected parameters, some have many others few).

A shameless plug for BBMap suite (bbmap.sh specifically) as a generalist aligner.

Answer 1

[dec986]

Through my own experimentation, STAR gives the best balance between selectivity and sensitivity. However, the mismatch rate is a little higher than HISAT2. I don't think that this is significant. HISAT2 comes as 2nd place, as it gets far fewer reads (sometimes tens of thousands with similar settings).

However, I am waiting with baited breath for an alt-aware aligner.

[Lars]

Short questions for my understanding: How do you evaluate selectivity and sensitivity? What is your gold standard for this statement? Why is a higher mismatch rate a disadvantage? Maybe the aligner is more sensitive by allowing more errors? And why are tens of thousands fewer reads bad? Maybe they are false positives? I always have problems with evaluating mapping results of dfferent aligners.

[dec986]

For my particular purpose, I downloaded GSE96075 and compared STAR against HISAT2. STAR's extra reads could indeed be false positives, but I had no way of verifying that. I was satisfied because the read length was fairly long (~198 nucleotides) so the slightly higher mismatch mean was acceptable. The max mismatch rate was the same between HISAT2 and STAR, but this setting can also be changed. I also really like STAR's two-pass alignment, read-counting, and auto-cleaning the data which avoids the necessity of choosing between different aligners (the same pipeline is used by the Max Planck Institute).

[GenoMax]

BWA-MEM has been alt-aware for a number of years.

[WouterDeCoster]

Based on the other aligners mentioned by dec986 I assume he means an alt-aware aligner for RNA-seq.

[Ram]

Off topic: The phrase is bated breath, not baited :-)

Answer 2

[Hussain Ather]

Hisat2 has given me the best results overall looking at the efficiency and results.