People underestimate how excruciating it is to collect large sample-sizes of genetic data.
The database 23andMe has is so stupidly valuable because 1. the sequencing is already done (and paid for) and 2. They can follow up with surveys electronically.
This kind of voluntary ancestry service will probably be the only way we will ever be able to collect the millions of samples needed for powered WGS GWAS analysis.
I think the 23 and me data is not sequencing data but genotype data. So it only looks at a certain type of mutation in a limited set of ~500k known mutations. I may be wrong so please correct me if so. So you won't find as many rare mutations in this data, or non-SNP mutations. Also I don't think they have robust clinical data for all subjects, it's just self reported. Again I may be wrong, I haven't done a 23 and me
This is a big deal for drug development. Each drug basically targets one protein. So you need a genetic marker that has a large effect size and is well correlated with a clinical phenotype. Not having clinical data is a big issue here. Also, 23 and mes database is not designed to find large effect size mutations -- their advantage is scale, and I think they only measure known mutations, so they are good at picking up common mutations with small effect sizes in common disease. The depression study they did is a good example of this application
But that type of study is low value for drug dev. You want large effect sizes. So if you have a big dataset, you want to find rare mutations with large effect sizes that are linked to extreme phenotypes, not common mutations with low effect size linked to common phenotypes. Basically finding more PCSK9 type mutations. Having only genotype data rather than sequencing data really hurts here
That's why I much prefer something like the Regeneron Genetic Center to 23 and me. They get robust clinical data, do while exome sequencing, and collect data from targeted populations where signalnis easier to discover.
In fact GSK was part of the RGC consortium but dropped out, dunno why. This may be their "rebound" from that
Again my assumptions about 23andme may be off bc I haven't used their product
>> The Regeneron Genetics Center (RGC) has built one of the world’s most comprehensive genetics databases, pairing the sequenced exomes and de-identified electronic health records of more than 300,000 people so far. https://www.regeneron.com/genetics-center
https://en.m.wikipedia.org/wiki/23andMe - users 5 million
My understanding is that you are saying 23andme's 5 million records are less useful for drug discovery than 300,000 records with
1. more detailed / reliable medical records 2. they do the wrong type of genetic analysis (which i don't fully follow)
Could you expand more? Can they re-use their genetic material if it was all stuck in a fridge and do the "better" sequencing?
Specifically, that the FDA prohibited them from proactivitely notifying users about potentially dangerous mutations. Looks like they've gotten approval for a few breast cancer risk markers now, but that's well short of what they could do.
If you're notifying users about a drug discovery program... the FDA can't gripe.
My argument: If you are exploring genomic datasets to find new potential drug targets, then what you really want as the output are single genes that are very strongly associated with dramatic phenotypes. Drugs generally only hit one molecule, so you want a monogenic vs polygenic signal, and drugs can't hit every molecule of a given type in the body, so you want a large effect size (if you only hit 10% of the target you still want a meaningful clinical effect)
If you don't have clinical data, your genetic data is Not super valuable for target discovery. You can't correlate genetics to clinical outcomes. The more clinical data you have, the more valuable your dataset, because you can uncover more gene-phenotype correlations. So you need high quality medical records, lab data, etc. if you just have self reported data on a few diseases, you'd miss all sorts of potentially useful signals
The type of genetic analysis is also super important. 23 and me does "genotyping": they have a chip with like 500k-1M molecular probes. Each probe is designed to detect a specific "SNP" mutation, i.e. A mutation where one DNA "letter" is different. So this doesn't pick up other non-SNP mutations but that's not as important. What's more important is that there are like 10M+ (edit: prev said 360M) known SNPs and prob a lot more unknown ones. So with 23andme you are only exploring a small part of the genome
And this part of the genome is fairly well explored. While next gen sequencing is a newer tech, gene chips (what 23andme uses) have been around longer. Most common mutations have been studied. And 23andme is just studying those common mutations but at a larger scale
However larger scale isn't necessarily that great for target discovery. With larger database you can pick up 1) more relationships with small effect size and 2) more rare relationships with large effect size. Except 23andme is using a gene chip that detects mostly common mutations, and bc they have limited clinical data they will mostly have common clinical condistion in their dataset. So you can really just pick up a lot of common mutations with significant but not meaningful relationship w outcomes
If you use exome sequencing like RGC, you get much richer coverage of the protein coding genome than 23andme. So you can pick up rare mutations. And you have more clinical data (arguably having more clinical data per genome is more valuable than having more genomes), so you can pick up more gene-phenotype relationships. You need to scale your sample set so you can detect rare mutations, but do you need 5M people? The more the better but RGC has already yielded some promising targets w it's smaller dataset
RGC is also smart and targeted about the kind of patients they recruit. So there is less noise and more signal, so you don't need as many patients. For example they look at fairly genetically homogenous "founder populations" that have less background genetic variability. Like the Icelandic population -- PCSK9 was discerned by analyzing this pop
PCSK9 gene is a classic example: mutations in this gene are very strongly associated with extreme levels of LDL cholesterol. And the relationship works for both gain of function mutations and loss of function mutations, and the causality can be validated experimentally.
The effect is dramatic: patients with loss of function mutations in PSCK9 have like 10% of the normal level of LDL cholesterol
And it's a monogenic trait: you can get these extreme LDL levels just by modifying PCSK9
So this is a great target assuming you can design a molecule to "block" it (you can). You can create an antibody that can basically have the same effect as the mutation (keeping PCSK9 from doing its job) but on a smaller and less durable scale.
The drug worked at lowering LDL cholesterol. It has had mixed commercial success for a variety of reasons unrelated to its effectiveness of lowering cholesterol
So this was really the first drug discovered based on large scale genomics. Regeneron developed one of the two PCSK9 inhibitors on the market. They purpose-built RGC to find more of these
I presume that GSK thinks they can do more detailed follow up medical questionnaires on the 5m existing users than just any random sample.
But will the genetic material still be viable for retesting with the more advanced chips?
Also it would cost maybe ~$500 / person to do whole exome sequencing, so doing this at scale is quite costly. Thats another reason you need to be targeted when recruiting populations for sequencing, it is so expensive. also RGC spent a lot of money on a high throughout, heavily automated sequencing center, so they can sequence more cost effectively than most. Even the UK Biobank genomic project is using RGC to sequence their patients
[#] insert your high quality long form news source here.
That's what the goal of https://allofus.nih.gov/ is. To collect data from 1 million people in the US.