How Accurate Is Liftover Utility For Mapping 1000 Genomes And Dbsnp 132 From Hg19 To Hg18?

Question

[Sam]

Hi,

I've been seeing mixed responses or unclear responses to this question on the forums so I was wondering if anyone might be able to clarify the matter. My understanding is that the liftover utility is generally a good tool for mapping between builds, yet I've seen it mentioned on the forums that mapping dbSNP 132 from hg19 to hg18 coordinates as well as the latest 1000 genomes release for hg19 to hg18 is not so trivial. Can anyone comment on their experience with using liftover to map variant annotations from hg19 to hg18 coordinate space? How accurate is it? For those regions producing innacurate results...does anyone have any advice to overcome the difficulty?

Thanks!

[Sam]

Just found this: ftp://ftp.ensembl.org/pub/misc-scripts/Assembly_mapper_1.0/

Anyone had any success with it? I'm going to be trying it out. Will post my experience with it

Answer 1

[Sean Davis]

I am not answering the question directly, but as a simple exercise, I used galaxy to do the following:

1. Get the snp130 data for chromosome 22 (hg18); 312,239 regions were included
2. Use the liftover tool to take data from (1) to hg19; 272,414 regions were successfully mapped to hg19
3. Map the output of (2) back to hg18; all regions from step (2) mapped back to hg18 successfully
4. Overlap regions from (3) with regions from (1); all regions in (3) were accounted for in 1

So, at least for a simple round-trip, it does not appear that any "errors" were introduced since all double-lifted-over snps were included in the original dataset, but there was a dropout of 39,825 (13%) of original snps.

Answer 2

[Shigeta]

I have done large liftovers from hg18 to hg19 and it depends.

If the SNP is well behaved and so is the sequence (low GC region, non repetitive regions) then you will find it works rather well. This is probably true for most of dbSNP.

On the other hand if you have a copy of dbSNP130 you don't need it - 98% of the SNPs will be in both collections :)

If you don't have it on hand I can probably get you a copy we have.