What are good settings for filtering VCF files?

Question

[devenvyas]

I followed the samtools/bcfutils/vcfutils pathway followed here to convert a set of human Hg19-aligned BAM files into a set of raw VCF files. I then got vcftools to filter down to just autosomal SNPs. These are really, really, really low-coverage genomes (they were enriched for NRY and/or mtDNA, and I am just trying to make use of the "leftovers")

Now I have the data I want, but I am trying to found out what of it is actually usable. I was wondering what are good filtering parameters for tossing/keeping human SNPs (or where can I find said parameters)? Thanks!

-Deven

Answer 1

[Ashutosh Pandey]

This is what I use. I generally change them depending on the study. But more or less this is close to what everyone uses.

• MinDP (Minimum read depth): 5 (Indels) and 3 (SNPs)
• MaxDP (Maximum read depth): You have a low coverage data, so I would set it to 100. Normally it is 3 times the average coverage.
• BaseQualBias (Minimum p-value for baseQ bias): 0
• MinMQ (Minimum RMS mapping quality for SNPs): 20 or 30 (to be more stringent)

• Qual (Minimum value of QUAL field): 15 or 20

• StrandBias (Minimum p-value for strand bias): 0.0001

• EndDistBias (Minimum p-value for end distance bias): 0.0001
• MapQualBias (Minimum p-value for mapQ bias): 0

• VBD (Minimum Variant Distance Bias): 0 (More relevant to RNA-seq reads)

• GapWin (Window size for filtering adjacent gaps): 30 bp

• SnpGap (SNP within INT bp around a gap to be filtered): 20 bp

• SNPcluster (number of snps within a region): I usually drop all the snps if there are more than 3 snps within 10 bp.

[Sheila]

@Ashutosh -

could you provide reasoning as to why those are the thresholds you typically use? It would be helpful so researchers can understand the parameters better! Thanks!

[Kevin Blighe]

You are right to question this - indeed, there are absolutely no standards for these filtering criteria. Take a look at my take on DP alone: A: DP in VCF files?

[devenvyas]

I know vcftools can filter based on DP/Qual, do you have any recommendations on what to use to do the other filtering? Thanks!

[poisonAlien]

This one does almost everything that's mentioned above.

[Ashutosh Pandey]

I have my own python script. If you know python you can modify it for your use. OR you can use vcf-tools "annotate" feature. I think the second option will be much better.