VCF + Phylogenetic tree : how to calculate the distance between two set of samples

Question

[Pierre Lindenbaum]

I'm trying to build a simple tool that would build a phylogenetic tree from a VCF.

My aim is to verify that the related + sequenced individuals are close in the phylogenetic tree.

My current code works with a small set of related individuals but it fails when some extra individuals are added. Current algorithm:

• a genotype is an enumeration Gtype: HOM_REF, HOM_ALT, HET . 'N/A' is converted to HOM_REF.
• I'm looking iteratively for the 'nodes' having the smallest distance. A node is a set of samples. The algorithm starts with all the possible pairs of samples.
• For a pair of samples
  • AA vs AA: = distance=0
  • AA vs AB: = distance=5
  • AA vs BB: = distance=15
• This is where I'm looking for the right algorithm : when merging two samples to create a 'merged node' how should I calculate the distance ?

My current algorithm for a position in the VCF is:

dist=0;
for(Gtype g1: allGenotypes(node1))
  for(Gtype g2: allGenotypes(node2))
      d += distance(g1,g2)
d /= ( countGenotypes(node1)+countGenotypes(node2))

What would be the best way to 'score' the distance between two sets of samples for a given position in the VCF?

Answer 1

[Jeremy Leipzig]

I would also look at how vcftools calculates relatedness

https://github.com/shantanusharma/vcftools/blob/b762dd590bdd4b4fb964385747ea382e2978d855/trunk/cpp/variant_file_output.cpp#L4061-L4174

Answer 2

[David W]

It's not quite clear to me what makes a "merged" node, do you mean a group that is already "joined" as each other's closest relatives?

If so this seems like a classic hierarchical clustering problem. You could use UPGMA or neighbor-joining to create your tree from a distance matrix.

Answer 3

[Zev.Kronenberg]

I would use pairwise identity by state for the distance matrix.