assemble resequencing data to multiple references

[ajthomas]

I have some targeted genomic resequencing data I'm trying to assemble. The genomic region is highly variable, and I have reference sequences from 8 different haplotypes I can use. The 8 haplotypes differ considerably from one another. So far, I've only been able to assemble the data to one of the references at a time, but I would like to find a tool that would allow me to use all 8 concurrently and assembly contigs based on best match.

Does anyone know of an alignment tool that can use multiple reference sequences of the same genomic region simultaneously? Or, can anyone suggest a better approach to analyzing this data?

[jimmybee]

Are you doing a de-novo assembly of the data? Or reference-based assembly?

[ajthomas]

Reference-based. As I said, I have reference sequences for 8 different haplotypes that differ significantly from one another. Rather than randomly picking 1 of the 8 to use as a reference, I would prefer to use all 8 to align as many reads as possible.

[xied75]

How about join those 8 together as one reference?

[ajthomas]

Wouldn't that just confuse the software? It would then have potentially as many as 8 sites to which it could align a given read, instead of just one. I think that would be comparable to just adding all 8 sequences individually as references because that would force the software to pick one of them to align each sequence to.

I was thinking about just creating a consensus sequence from the 8 and using that for the reference, but I would like to keep the 8 sequences separate for comparison purposes later.

[xied75]

Yes it'll confuse the aligner a lot! For the duplicate area, it will need to pick one from the 8. But for variable area, it will report the highest scored one.

For what you want, I guess no aligner is designed for that purpose (my guess). Because instead of creating an aligner that can align to 8 ref at the same time, why not start 8 instance each targetting one ref? It was only lines of script away from you.

[jimmybee]

Reference-based. As I said, I have reference sequences for 8 different haplotypes that differ significantly from one another. Rather than randomly picking 1 of the 8 to use as a reference, I would prefer to use all 8 to align as many reads as possible.

How about assembling it de-novo, match your region of interest with BLAST and then align each haplotype using a multiple sequence aligner like CLUSTALW etc? This would remove any bias that the reference sequence would have on your reference-based assembly

[ajthomas]

That's not a bad idea. I'll have to think about that some more to decide whether I could make that work. I could do the de novo assembly, then BLAST the resulting contigs against a local BLAST database of the reference sequences, then use that information to map those matches back to the original sequences. Better yet, I could map them back to an alignment of the 8.

I've tried aligning these sequences to create a consensus, but haven't had any luck yet. They're about 5 megabases each, and attempting to align even two of them brings my computer to its knees. (HP Z800 with 2 6-core CPUs and 48 GB RAM) I'm looking into using the university's high performance computing cluster core next. I hope that can do it.