Aligners, aligners, aligners

[dawe]

Hi all,
I'm working with Illumina GAII and it's has been clear for months that eland is no more the fastest (nor the best) aligner available (and I wonder why Illumina doesn't include an alternative tool...).
BTW, there are many aligners out there and I'm puzzled on which one should I use. I've read the paper by Palimeri & Schlötterer which is only telling a part of the whole story (some aligners are not mentioned, options for tested aligners may not be optimal...). In the end I've been focusing mainly on bowtie and bwa, mainly for speed, then for SAM compliance (and I'm only assuming SAM will be a standard for all aligners someday).
I know there are many developers reading this forum, and probably they will tell that their software is better... what about the others? What would you choose and why? Also, who has played with all the CLI options available to see how results vary?
d

[lh3]

I would recommend maq, bwa, bowtie, novoalign and soap2. So far as I know, all these are under heavy daily uses by people from different centers. You might also find the following two pages (here and here) interesting. They are quite lengthy and one page is less related to the topic, though.

[Cole Trapnell]

I second lh3's list - the five tools mentioned produce results that are in broad agreement with each other, despite using radically different algorithms. Each has different strengths and weaknesses, but provided that you pick one that is compatible for your data (for example Bowtie doesn't support Colorspace yet), you should get good results for your downstream analysis.

[nilshomer]

I would recommend MAQ, BFAST and BWA. I recommend MAQ because it was the early standard and a compromise between ease-of-use and power. I recommend BFAST because I am the author as well as many other documented reasons. I recommend BWA because it is fast, good for low-error data, as well as small (<3bp) indels.

Note: I dislike against any aligner that specifically ignores insertions and deletions (not just the small ones), which are quickly becoming an important mutation mechanism in cancer (for example). This makes me biased against aligners that do not use a Smith-Waterman local alignment to detect errors/variants (especially for ABI color space).

Specify up front what you want to find (if only perfect matching reads with no variants, that's fine), then choose your aligner. Of course, aligners tuned to find only reads with up to one error will be faster and easier to use (so use them in those instances)!

[ohofmann]

The 'robust' list posted above is a good starting point, with another vote for Novoalign. Completely depends on what your aims are -- we've added Mosaik to the list because we needed support for IUPAC codes, but also started with some of the algorithms that come prepackaged with EC2/Cloud scripts just for convenience.

[der_eiskern]

novoalign and bfast appear to be quite similar to me. nils, care to comment on the differences?

[BaCh]

Hi all,
BTW, there are many aligners out there and I'm puzzled on which one should I use.

The list given above is a good start for aligners.

Depending on your needs, you might also want to look into assemblers. E.g.: using Solexas to polish a Sanger or 454 assembly. Programs that map Solexas to existing assemblies ideally dump the result in a format you can then directly use in finishing programs to manually adjust the last problems remaining.

B.