Local Assembly For Verification Of Indels

Question

[lyfsa]

Hi There,

I am trying to further verify indels detected from variant caller. For this, I extracted all the bam files reads, where indels were detected, run local assembly and realign the contigs from assembler back to reference to get the exact coordinates. My assumption is that, there should be gap in the regions where indels are detected. Can somebody provide an insight in this?

Thanks

Answer 1

[JC]

Actually, in my opinion, you are doing it right, I did the same sometime before with some Illumina paired-end data, I tested Velvet, ABySS and SOAPdenovo, when the region is small, all programs works perfectly, when you have a complex mix, ABySS can resolve better but SOAP produces longer contigs.

[lyfsa]

Could you also tell me how velvet works for single end reads or unmapped reads in BAM files? I am quite skeptical about these reads.

[JC]

the perfomance depends a lot in your library size, deep, quality and k-mer selected, so you gonna need many tries.

[lyfsa]

Thank you..you information is of great help...

[lyfsa]

After running the velvet and realigning them back to reference using mummer, I get several numbers of coordinates with 100% match with the ref sequence but I cannot make out long sequence out of that. I now have numbers of reads with many gaps in between. Can I say that if contig regions are overlapping with indels region, than those overlapping indels regions are false? or this is not the case. Sorry for asking so many questions but I would be grateful if you could help me.

[JC]

as I said, the contig size and quality depends in many factor, try to change your kmer-size in velvet to see if you can get longer sequences

Answer 2

[Jorge Amigo]

one quick and easy way of checking this while comparing your variant calling results with a reference software would be to use GATK, since in the suggested pre-analysis steps it includes a local realignment one around indels. check its best practices page to have this process better described.

[lyfsa]

I went through that link. But it looks like I cannot validate indels result from different caller??

[Jorge Amigo]

my suggestion was either to use the full GATK pipeline described on the link above and compare the final variant calls with the previously obtained, or to use just the IndelRealign walker for your convenience, since it already provides local realignment of reads around indels detected and also around know indels you may provide (such as all known indels from 1000genomes, for instance).