Hacker News Reader

Comment by dannykwells | Hacker News Reader

This comment is not correct as to what 23andMe is doing with the data.

The goal is to identify targets for new drugs, and it is known that drugs based on specific genes or mutations often have the best chance of success - look up, for example PCSK9 inhibitors. 23andMe, along with having genetic information, have very, very detailed phenotypic information on the people they sequence, which they collect in the form of surveys. This in turn lets them do GWAS-like analyses better than anyone else on earth. These GWAS hits can in turn go through a pipeline to find those genes which might confer protection, or susceptibility, to a particular disease. Then, it's up to GSK to make a molecule to mimic or inhibit that effect.

What should be clear here is that 23andMe is NOT giving GSK access to their data. They are giving them leads.

Source: work in research R&D in biopharma (with a particular interest in genetics)

View on HN

Sorry, the blanket dismissal is not entirely correct. Companies like 23andme are indeed trying to repurpose their data to efficiently enroll participants in trials, as the parent post suggests. While it's not the focus of the GSK partnership as publicly reported, it is absolutely part of the strategy in this market, and arguably more likely to succeed in the long run. As far as target discovery: yes, there are examples of success with GWAS data (though probably more with rare variants identified in sequencing), but 23andme phenotypic data is rather craptastic vs traditional studies or biobanks. No doubt they'll find things, but you're overstating the power of their data. Source: work in drug discovery in academia; I've been pitched by companies like this and worked with a few.

Agreed re: target discovery. Most of the 23andMe data is not much better than what is already publicly available to researchers via initiatives like the UK Biobank. 23andMe uses the Illumina Global Screening array which has significantly worse output (in terms of number of SNPs) than the 500k patients genotyped in the UK Biobank. Furthermore these genotyping based assays make it impossible to detect rare variants that have never been seen before and may be protective. The data being generated by Helix is probably much better for target discovery, but still not competitive with private datasets constructed by companies like Regeneron.

I don't know exactly what phenotypic data 23andme gets but if it's through surveys then it probably isn't as good as the clinical data that, say, regeneron gets for its genetic center through partnerships with healthcare providers. Also 23andme's database is large but presumably much of their data is from young healthy people -- I don't know for sure but it seems a reasonable assumption. So it wouldn't be as useful in studying disease as the overall size of the dataset seems to suggest. It also isn't as useful as a company like decode or Genomics medicine Ireland that study more homogenous populations. And GWAS have been done for many years so much of the low hanging fruit is picked

23andmes database's best advantage is probably studying things like genetic basis of mild to moderate anxiety and depression as those diseases are probably more prevalent in their population than others. Plus those are complex polygenic traits so each individual gene would have a small effect size, so a big database like 23andmes would be useful

However that isn't great for drug discovery because drugs only hit one target. A bunch of genes with small effect sizes won't support much of a drug dev program

I may be wrong as I'm not familiar w 23andmes database but these are my deductions from what I do know

Reply on news.ycombinator.com