question on predicting the function of a variant

[kjaja]

Hi All,

I have a question in prediction of function for mutations. Is it better to use Polyphen2 or SIFT score. I came across a mutation that has a low SIFT score (0.01) but when I used Polyphen, it was predicted to be “benign”. Is it better to go with SIFT or Polyphen2 and how do they differ.

I appreciate your help

[Heisman]

This is something that has not been solved well. If you use ANNOVAR to annotate your variants, it will query against multiple databases (newest version does PolyPhen2, SIFT, LRT, PhyloP, MutationTaster, all according to the dbNSFP paper http://onlinelibrary.wiley.com/doi/1...21517/abstract), as well as the GERP++ score and it will tell you if the variant is present in dbSNP135, as well as the frequency it is present in the 1000 genomes project and the Exome Variant Server.

Now, how do you interpret that? I once for fun took all of the SNPs in dbSNP135 that were stated to be confirmed damaging or confirmed non-damaging and ran them through ANNOVAR to compare all of them. Pretty much none of the metrics showed a big enough distinction to be useful, with the exception of PolyPhen2 and especially MutationTaster. So I would say those are the best. HOWEVER, I do not know if the SNPs I was looking at were included in the training sets for those two algorithms; if they were than this was useless.

So, I would trust PolyPhen2 over SIFT, but I wouldn't have much confidence in either.

[dlgoode]

You might be interested in this paper that recently appeared in Bioinformatics:

Performance of computational tools in evaluating the functional impact of laboratory-induced amino acid mutations

http://bioinformatics.oxfordjournals...ef-corresp-1-1