Hacker News Reader

Comment by gingabriska | Hacker News Reader

Can we use 23andme data to see if we've this gene?

It really depends. In this article, they describe a complex genetic architecture, with copy number variations (the entire gene exists in multiple copies along the genome) as well as deletions and insertions. 23andme data, being based on microarrays, probably gives a misleading read on this because the complex genetic architecture is not encoded into the microarray.

I generally recommend against reading primary liuterature or the PR around it and immediately checking genetic tests to see if you "have this disorder". IN complex diseases, many genotypes have low penetrance (IE, the genotype variable is only weakly correlated with the phenotype label), and diseases like this are complex, polygenic, and any individual mutations only partially explain the visible phenotype.

Perhaps a stupid question (as I am neither very educated on the topic of genetics nor have I used 23andme myself), but isn't there a way to download the dump of your raw genome data (that is compatible with all the other genetic analysis tools out there) from 23andme that would allow you to find it?

Btw fully agreed with you on the points of complexity and regular people misinterpreting their results following the PR hype.

When you download that dump from 23andme, unless you signed up for whole genome sequencing, and even then, the dump will not include high resolution data which would allow you to analyze these results in the context of the paper.

If you did WGS (whole genome sequencing), you might be able to take the VCF files (or the FASTQ) and, hopefully, their mapping process did an accurate job computing your individual genotype. However, in cases like this with CNVs and indels, I wouldn't really trust the assembly. MOst of the tools that we built to produce genotypes get a lot of the details of the mapping or assembly wrong. You could do a lot better if you had your genome sequencing using both long and short read tech, and used the long reads to make a reliable scaffold and then mapped the short reads, but with CNVs, much of that is still guesswork.

Ah, that clarifies it, thank you. I didn't realize that the raw dumps on 23andme would differ depending on which version of the product you paid for, as well as the inherent imprecision of those.

If you want a high quality genome (ilumina UYG) to start with, it's https://my.pgp-hms.org/profile/hu80855C which includes the original BAM files (so you can try your own mapping) as well as called variants (conveniently[?] segregated into SNPs and indels).

I would expect "everything" you need (except for long reads) is here for you to determine if the owner of that genotype (I can vouch for them...) has the genetic alterations described in this article (https://www.nature.com/articles/s41467-019-11119-7) after performing some sort of magical rat/human remapping (tricky) or referencing off of https://www.omim.org/entry/615656

I would be impressed if somebody could actually make a clinically actionable recommendation for this genome, as Illumina described it as have zero known risk.

> Most of the tools that we built to produce genotypes get a lot of the details of the mapping or assembly wrong.

Can't you use the existing reference human genome as a scaffold?

not if somebody has a complex genetic architecture with differences in areas of high self-similarity, or copy number. The reference lacks areas (and has areas) that are in or completely missing in other human populations. It took people decades to acknowledge this deficiency (which is due almost entirely to using a very limited sampling of humans for the first references).

Reply on news.ycombinator.com