Suggestions for aligners?

[kga1978]

Hi all,

We have been trying out various aligners over the last many months, but I wanted to get some input from the community on suggested aligners. The genomes we have are as follows:

Genome:
Viral (RNA-based)
10kb genome
High divergence (5-10% normal, sometimes 15%)

Considerations:
We don't care about memory (small genome)
Aligner must be able to deal with divergence and gaps (few)
We don't care about (super) speed
Preferably scalable and 'pipeable' (samtools, picard, GATK, etc.).

Goals:
Consensus sequence for 'standard' phylo studies
High-coverage (>150x) for intra-host SNP detection

We mostly use illumina 50bp SE sequencing, but also have 100bp PE and ~500bp 454 sequencing. We have been testing multiple different aligners and have focused on the following:

BFAST
BWA
SSAHA2
NovoAlign
Mosaik (mostly v1, but v2 just came out)
Stampy

So far we have had most success with Mosaik and NovoAlign in terms of specificity and sensitivity on the Illumina platform. For 454 we have only used Mosaik for now (which works well - but homopolymer and CAFIE errors have to be cleaned up manually). For these two tools we generally use a hash-size of 6 and a divergence of 0.1. For Mosaik we also specify an 'act' threshold of 10. We have tweaked several other parameters, but have found them to influence the alignments very minimally.

I was wondering if anybody would have further insights or suggestions? We are currently scaling up production so any suggestions and comments would be most welcome.

[kga1978]

Great question! I do wonder if anybody would have some input...?

[mchaisso]

Try bwa-sw, it is shown to work well for 454 reads and divergent genomes.

[gsgs]

I've been using MAFFT for influenza

http://mafft.cbrc.jp/alignment/server/index.html

but recently figured out that it is overkill for most
practical problems, when you have a common ancestor-sequence,
a referrence that you can align to.

I've built a list of ~200 index-sequences to which I can quickly
compare new flu- lists (e.g. genbank updates) without alignment,
to which group they belong.
Then I group the sequences accordingly and do index-base-alignment
for each group separately.

Or I align first without gaps (easy+fast) and then filter out (discard)
sequences that are distant, didn't align well.
They may have insertions,deletions or sequencing errors.
These ~1% of sequences are then omitted or aligned with
MAFFT (slower) in a 2nd step.

I did this for e.g. for all avian PB2 or such, 10000 sequences
of length 2280, max.~400 mutations