Using full haplotypes in BAM files

[RubenMarchau]

Hey,

for my research I am looking to use haplotype frequency in outbreak reconstruction to incorporate within-host variation.
In the phyloscanner article (Wymant et al., 2017), there was mentioned that haplotype reconstruction was considered but did not yield satisfactory results.

Therefore I was wondering, could phyloscanner use haplotype frequencies by using BAM files where

• each read is the length of the entire genome (i.e., one complete haplotype),
• identical reads represent the same haplotype repeated according to its frequency

Would phyloscanner be able to handle this type of input?

Thanks!

[mdhall272]

Yes - that's been done, more or less, in http://elifesciences.org/articles/46402 (they weren't reads per se, but the principle is the same).

One thing that changes is that if your window is the whole genome, you will get only one ML tree, and thus there is no scope for quantifying uncertainly with different windows. So I would either apply phyloscanner to a set of bootstraps, or to a Bayesian posterior set (e.g. from MrBayes). That's honestly rather more rigorous, not less!

(Note that we rejected haplotype reconstruction in the original paper based on the sequencing we had at the time. With longer reads, it might work better; but if your read is the whole genome, that's better still.)

[ChrisHIV]

The simplest way of calling phyloscanner_make_trees.py (see the manual or repository home page example) will work out of the box for what you suggest above: merging each set of identical reads into a single sequence with an associated count, what you're calling a haplotype. Don't forget about sequencing error though: a single base change will cause a read to go into a separate unique sequence or haplotype. phyloscanner_make_trees.py has the options --merging-threshold-a and --merging-threshold-b for merging distinct sequences if they differ by a small specified number of base changes. Or of course you could use phyloscanner just to merge and count identical reads, and then do your own downstream processing to merge/cluster unique sequences that are so similar that you suspect it's the same haplotype with sequencing error.