I’m looking to use Ti/Tv as a quality evaluation metric for human WGS variant calls. My current thinking is to calculate it using SNVs that:
- Fall within the 1000 Genomes strict accessibility mask
- Are biallelic only
When I apply both filters, I get a Ti/Tv of ~2.1, which seems reasonable for genome-wide human variation.
My rationale for the accessibility mask is that I think low-accessibility regions may contain mapping/sequencing artifacts that distort Ti/Tv. For restricting to biallelic SNVs, I noticed that multiallelic sites have a much lower Ti/Tv (around ~0.7 in my data). That made me wonder whether multiallelic sites inherently skew Ti/Tv because each reference base has only one possible transition but two possible transversions.
So I’m mainly wondering:
- Is restricting Ti/Tv calculations to accessible regions and biallelic SNVs considered standard practice?
- Do people typically include or exclude multiallelic SNVs when reporting Ti/Tv?
Thanks!
•
written
by
curious
1 answer
Your multiallelic intuition is right, and the bound is tighter than you'd think. Each reference base has exactly one possible transition and two transversions, so a triallelic site can only give Ti/Tv of 1.0 or 0 -- a set of them can't exceed 1.0 no matter how clean the data is. Random ALTs would give 0.5, so 0.7 is about what real data with a genuine transition bias should look like. It's arithmetic, not a quality signal.
Both your filters are standard, yes. The thing I'd add is splitting by dbSNP membership and reporting known and novel separately. Novel always runs lower, but a big drop there tells you far more about false positives than a pooled number does, since known variants dominate the count anyway. VariantEval does that split for you.
2.1 genome-wide is the right ballpark too. Exome runs 3.0-3.3 and people sometimes panic comparing a WGS number against a remembered exome figure.
Log in to answer this question.
