You can tell a surprising amount about the genome from a subset of SNPs. Most germline variants occur in groups, so you only need to measure one position to know the values for that group. The variants that are chosen for the array based tests are carefully chosen to give you the most amount of information possible. If you’re interested in this, look up “genome imputation”.

For most of the tests that 23andme would want to run on a population, like risk assessment, you don’t get much more information from whole genome vs arrays. And the extra data (and noise) you get from whole genome can make association studies significantly more difficult.

Also with whole genomes, the extra cost in sequencing, storage, and processing time (all of which are significantly higher) wouldn’t be worth it from a cost/benefit point of view. Remember, most of the data is redundant or common between people, so full sequencing is really only beneficial in specific scenarios.

The issue with SNPs is that they are generally common in the population, and a variant that is common will rarely have a big effect. If we look at GWAS studies, most of them do not explain much of the variance of a phenotype. It either is due to environment/epigenetics or we are not using the right data. Part of it is of course the former, but I'm betting on the latter. I'm not a researcher, so I could of course be very wrong.

There is a difference in scale. For the research 23andme wants to do, they need to identify variants that occur in a significant part of the population (~1% or so). They need to do this because any potential drug/treatment would have to be developed for a big enough patient pool. Large population level genome studies like 1000 genomes and NIH All of Us studies are all aimed at finding those ~1% variants.

But if you’re looking for what is going on in an individual, then yes… SNP chip data isn’t as informative as whole genome (or exome) sequencing.

If most data is common between people, then storing the diffs should be cheap.

It still costs a lot to process and until it’s all aligned, you don’t know which parts are common. So, even if you have smaller long term storage, whole genome data still has significantly more costs to work with (at least an order of magnitude, if not two).

And you store more information than just the raw sequence (quality scores, multiple reads, etc).

I've asked a virus biologist, and he said there's actually not that much interesting in full genome for them. That "small subset" is all he needs.

But it's a crucial subset. It's not that WGS is that expensive, but what can you do with it that cant be done with 23andme?

Correct me if I'm wrong. The issue with SNPs is that they are common in the population, and hence they tend to have small effects, because if a variant is common and gives the carrier a big advantage, the variant will over time get absorbed by most of the population, and the same applies the opposite way. So the interesting variants are those that rare, because they are the ones that often cause genes to malfunction.

point mutations?

That's what SNPs are. Specifically, point mutations that are associated with known outcomes (ie diseases) or with group differences.

Point mutations occurring outside SNPs are almost by definition not yet useful. If we know what the gene does or is related to, it's an SNP.

You second paragraph is wrong. SNPs are interesting because they exist in a significant part of the population.

But a new mutation never reported before can as dangerous as you want. A case we found: a woman already developed an illness in her adult life, lets supose it was renal failure. We suspected a genetic root, so we sequenced her exome, founding a never reported mutation that frameshifted a gene previously associated with renal failures. We quickly panel-sequenced her parents and children to check if they had the mutation, which none of them had. So the woman had a de novo mutation in heterozygosis that their little kids were lucky enough to avoid. Imagine the relief. Not useful? I disagree.

but are the ones that are not already in the covered snps interesting ?

23andMe only has an overlap of ~30k with the ~500k (?) ClinVar variants. So less than 10% is covered.