When should you left-align INDELs (and why?)

Question

[QVINTVS_FABIVS_MAXIMVS]

Say I have two VCFs with 100 samples in each file. Each VCF was joint-called separately and now I want to merge the variant calls.

Do I need to left-align the INDELs in the merged VCF? I've used bcftools norm in the past and got odd results. It seems that vt is a better tool for this.

Is left-aligning only useful for common variants? If I'm interested in rare variants (<0.5% AF) would left-alignment actually matter?

Thanks

Here's an example of bcftools norm

Original VCF

chr7    157009949       .       AGCGGCGGCGGCG   AGCGGCGGCGGCGGCGGCG,A,AGCGGCGGCGGCGGCGGCGGCG,AGCGGCGGCG,AGCGGCGGCGGCGGCGGCGGCGGCG,AGCGGCGGCGGCGGCG

---

Left-Aligned VCF (with multiallelics split into biallelic calls)

chr7    157009949       .       A       AGCGGCGGCG  
chr7    157009949       .       A       AGCGGCGGCGGCG
chr7    157009949       .       A       AGCGGCG
chr7    157009949       .       A       AGCG
chr7    157009949       .       A       AGCGGCGGCGGCGGCG   
chr7    157009949       .       AGCG    A
chr7    157009949       .       AGCGGCG A
chr7    157009949       .       AGCGGCGGCGGCG   A

---

[finswimmer]

Hello,

could you please give an example of an "odd result" of bcftools norm?

fin swimmer

[QVINTVS_FABIVS_MAXIMVS]

Edited the main query above.

Answer 1

[chrchang523]

Suppose one of your VCF files has a non-left-aligned insertion, represented as REF=AG, ALT=AGT, starting at position 99999, and another file has an insertion represented as REF=G, ALT=GT, starting at position 100000. If you don't left-align, these may not be recognized as the same variant, and downstream analysis will suffer.

[QVINTVS_FABIVS_MAXIMVS]

Thanks for the clear answer. Do you recommend vt or bcftools for normalization?

[chrchang523]

Either will work (as long as you aren't using a very old bcftools version). The latest bcftools should be faster, especially if compiled with "libdeflate".