Phased psuedo-kit on GEDmatch -- what accounts for huge drops in SNP count?

Question

I guess first I should make sure I understand how GEDmatch stores raw data -- I'd guess that during "tokenization", GEDmatch turns my raw data file into a format that it likes, i.e., perhaps replacing A, T, G, C with numbers or replacing names of loci with numbers, etc. And because raw data comes from different companies in different forms, this process "standardizes" the information GEDmatch is using. But I assume it is equivalent to having the information in my file -- i.e., a list of loci names, corresponding chromosome and position, and a pair of nucleotides, i.e., A, T, G, or C. Is this essentially correct?

I saw recently on another forum a description of what GEDmatch does to create a psuedo-kit phased between a parent and child:

"They are then simply taking the SNP differences between the child's kit and the parent's kit, removing those alleles of the tested parent that can be confirmed to be necessarily present in the other parent, removing any no-calls, and creating the new "pseudo kit" from those values...which will always have decidedly fewer SNPs than actual raw test results. "

Here's what I draw from this post:

As I understand it, a phased parent/child kit is a substitute for the child's kit. The idea is that the substitute will reveal matches to the child known to be on that parent's side. It sounds like data is stored as pairs of alleles, and when a single allele in the child does not match either allele of the corresponding pair in the given parent, it is attributed to the other parent and stripped out, and no-calls are stripped out as well.

Is this roughly what the phased kit will contain? Then here's the problem I can't wrap my brain around: I would guess less than half of the alleles would be removed since one allele in each pair for the child DID come from the given parent and sometimes both parents would have matching alleles by chance (Also I presume the no-call percentage is just a few percent).

Using GEDmatch Genesis, I phased my kit with my mother's (hers was uploaded to GEDmatch) and saw the number of tokens decrease from 450000 to 190000 in the phased pseudo-kit. Perhaps this has something to do with Genesis being in beta and the kits being on the different sites, but I've heard reports of SNP counts dropping almost 10-fold between a kit and a pseudo kit. And in my own case, the number of tokens used used for chromosome 1 dropped from 37969 to 0 after phasing!

What is wrong with my understanding and accounts for these huge drops in amounts of data?

Answer 1

GEDmatch does not disclose its methodology for phasing at GEDmatch Genesis. But if you and your mother tested at different companies, especially different companies that use different chips, then it could be that GEDmatch Genesis only includes SNPs that it can phase that are in both sets of raw data.

ISOGG has an Autosomal SNP comparison chart which I'm reproducing here.

You'll notice in the chart that 23andMe v5 tests 630,132 SNPs and FTDNA tests 698,179 SNPs. But the two chips only have 162,602 SNPs in common. No-calls on either side will also have to be excluded. Where there is only a parent SNP or only a child SNP and its allele values are the same, aka Homozygous (one of AA, CC, GG or TT) then the repeated value will be the phased value and the other reading isn't needed. GEDmatch Genesis could be doing all this and trying to get as many valid phased SNPs as possible.

Why chromosome 1 dropped to 0, I have no idea. Unfortunately, GEDmatch doesn't allow downloading of their Raw Data Files, so we can't see precisely what it created. Maybe you could run your kits and the phased kit through the GEDmatch Genesis DNA File Diagnostic Utility and see if comparing their summary data helps explain why your phased kit is different.