Whole Genome Alignment

[AdrianP]

Hello,

I have 6 sequences of about 7mb to align. The difference between them is just indels and SNPs.

MAUVE so far is not doing a very good job.

Anyone? I tried mafft but it never finishes.

[mchaisso]

Hello,

I have 6 sequences of about 7mb to align. The difference between them is just indels and SNPs.

MAUVE so far is not doing a very good job.

Anyone? I tried mafft but it never finishes.

Is there a way to quantify how it is not doing a good job?

[AdrianP]

Is there a way to quantify how it is not doing a good job?

In term of quantification, perhaps not. But when I zoom into the alignment, it becomes obvious that for some regions of any 1 isolate, it places them separately, with no alignment to any other 5 isolates, while the other 5 are aligned properly. And it does that a couple of times for different isolates in different regions.

These sequences are very very homologues. I know that because I obtained them by mapping reads of different isolates to a reference and regenerating the reference.

[winsettz]

An alternative way to frame this multiple sequence alignment problem might be to map these giant sequences to reference using bwa or novolign, then looking at it on tablet, IGV or sam/bam visualizer of choice?

If I understand your first two sentences, you mean to say:

-For any one isolate, there is one region of sole disagreement to reference, but all other isolates agree with reference.

-There are more than one region, and more than one isolate, with this behavior.

Is there anything special about the genomic content of the "regions" in question? Ambiguous N's? Sequence repeat? Mobile genetic element?

[AdrianP]

From illumina HiSeq of 6 isolates, I did denovo assemblies, since the reference currently available for this genome is draft and terrible.

I picked the best assembly of the 6 isolates, concatenated all contigs bigger than 10kb and used it as a reference to map the other 5 isolates to it and generate individual consensus.

I suppose I could take the other 5 new-references and map it on the 6th with bwa.

I am not sure about the details of the structure of the genome, i just know that the concatenated contigs are not scaffolds, so no stretches of NNN.

The purpose of this is to identify SNPs between the isolate that are 100% variant sequence (homozygous), and build a tree. Don't we all want a nice tree... sigh.

[winsettz]

The other possibility is to pick a best isolate, then call SNPs with samtools. Haven't played with SNP calling lately, but from C. Titus Brown:

http://ged.msu.edu/angus/tutorials-2..._tutorial.html

Edit: and from samtools themselves http://samtools.sourceforge.net/cns0.shtml

[AdrianP]

I thought about that, I can also use freebayes. The question however is, how do I use a vcf to construct a phylogeny? An alignment is necessary.

[lh3]

In term of quantification, perhaps not. But when I zoom into the alignment, it becomes obvious that for some regions of any 1 isolate, it places them separately, with no alignment to any other 5 isolates, while the other 5 are aligned properly. And it does that a couple of times for different isolates in different regions.

These sequences are very very homologues. I know that because I obtained them by mapping reads of different isolates to a reference and regenerating the reference.

It seems that the assembler is not doing a perfect job, or biologically some isolates are different. You should compare the short-read alignment and the assembly alignment to see which is the case.

[AdrianP]

Mapping reads back to contigs I see paired end reads supporting the contigs along the entire sequence of each contig. Not sure what else I could do to check the assembly.

I would be more interested in how to extract SNPs for vcf files and constructing a phylogeny with that. The problem is that it may find homozygous SNPs at different locations in different isolates, so it seems like a script is necessary to create an alignment.

Adrian

[lh3]

Map the reads to the reference genome to see if the lack of the contig alignment is real biologically or due to assembly problems.

You can call SNPs jointly, concatenate all variants in VCF and construct trees based on the resulting pseudo-sequence. Nonetheless, you won't get a meaningful branch length this way.