How To Filter Against Common Snp Database To Look For Rare Variants?

Question

[Bioscientist]

I've been using samtools/GATK to call for SNP/indels these days, and would like to filter my data against known common variants, to achieve rare events.

Some of the samtools results can be:

#CHROM    POS    ID    REF    ALT    
1    10177    .    ACCT    ACCCT

There could be several alternatives for both REF and ALT sequence. So when I compare my SNP/indel with common SNP database; should I just compare the position using bedtools, or should I look into the change of bases? For example, if in database, on position 1, it's a SNP changing from G to C; while my results indicate also at position 1,but changing from G to T. Then should I regard this as common SNP or rare SNP?

Thanks!

Answer 1

[Vikas Bansal]

First of all, I will annotate my SNPs by checking if they are present in dbSNP (same change). You can use tools for this, may be ANNOVAR and it can perform filtering also. I will not simply exclude if a SNP is present in database because may be that SNP can be important in your study. One way is to exclude SNPs which are present with more than 1% allele frequency in database.

EDIT:

From UCSC, you can find this information from here. It says-

Common SNPs(135) - SNPs with >= 1% minor allele frequency (MAF), mapping only once to reference assembly.

Flagged SNPs(135) - SNPs < 1% minor allele frequency (MAF) (or unknown), mapping only once to reference assembly, flagged in dbSnp as "clinically associated" -- not necessarily a risk allele!

Mult. SNPs(135) - SNPs mapping in more than one place on reference assembly.

All SNPs(135) - all SNPs from dbSNP mapping to reference assembly.

[Bioscientist]

so basically you mean we should consider the change of base, rather than simply the position, right? thx

[Vikas Bansal]

Yes, and also in downstream analysis, what kind of mutation is- missense, nonsense , frameshift etc.

[Bioscientist]

Also, the dbSNP, you mean all SNP, common SNP? There are two different database on UCSC, one "all SNP", one "common SNP"

[Vikas Bansal]

See my edit. Select accordingly.

Answer 2

[User 2005]

Here are a few guidelines :

1. Annotation against reference genome
2. Detection of non-synonymous SNP
3. Group SNP by gene of interest (associated with pathology) or gene ontology
4. if nothing comes out, do a chi-squared for each position found and see what is your treshold p-value
5. Manhattan plot / QQ Plot

I hope it helped :)