23andMe is exploring splitting the DNA company in two | Hacker News Reader

23andMe is exploring splitting the DNA company in two

bloomberg.com

12 pointsmfiguiere21 comments

23andMe has always been a shady company. It would be better for them to go out of business and all the data they collected on people be deleted. Of course they are looking for new ways to bypass any privacy laws to share peoples genetic info to the highest bidder.

Everyone I know regrets sending off their DNA to them for ancestry parlor tricks and other non-relevant information.

So... mitosis?

What about your toes?

Why sell off your customer's DNA once when you can sell it off twice?

> Wojcicki said one of the challenges is how to get customers spend for new products, like blood testing.

I can believe that some cancers could be detected early via nucleotides found in the blood.

However, Elizabeth Holmes might like a word here. Testing people's blood is a highly regulated activity. One does not simply jump into it.

There are already several companies selling liquid biopsy tests for early stage cancer screening. Basically they look for cell-free DNA containing mutations highly correlated with cancer. Galleri is probably the biggest player in this space.

https://www.galleri.com/

it says "In a clinical study, around 4 out of 10 people with this result were diagnosed with cancer after diagnostic testing."

so is my arithmetic correct, in that it's a 60% false positive rate?

That corresponds to the publication, https://www.thelancet.com/journals/lancet/article/PIIS0140-6... "A cancer signal was detected in 92 (1·4%) of 6621 participants with analysable results. 35 (38%) participants were diagnosed with cancer (true positives) and 57 (62%) had no cancer diagnosis (false positives)."

I am surprised that they were approved based on their study results, it will be interesting to see the results from PATHFINDER 2, which is running now.

If there a legal reason that the test shouldn't be approved? Although the false positive rate is high, it's ultimately up to the patient and their doctor about whether to order the test and what to do with the results. I do understand the concerns over false positives unnecessarily scaring patients and leading to wasteful additional diagnostic tests, but I don't think the FDA really accounts for that in the approval process.

Detecting early stage cancer in a liquid biopsy is extremely difficult so it's impressive that they made it work at all.

It's policy not law (the law give the FDA authority to set policy and makes decisions). And they definitely would consider the sensitivity and specificity and PPV of a test. TBH it's amazing it got approved. But the product leadership and founding team definitely knew how to play the FDA game.

Yes, early cancer tests are fantastic. However, tests need to have extremely good sensitivity and specificity for exactly the reason you said.

Assuming that you calculated that by subtracting 4 from 10, then divided the results by 10, and expressed that as a percentage, then yes, your arithmetic is correct.

It's just not the right arithmetic for computing the false positive rate. :-)

What you computed is the false discovery rate.

Let

  FP = number of false positives
  TP = number of true positives
  FN = number of false negative
  TN = number of true negative

What you computed is FP/(FP+TP). The false positive rate is FP/(FP + TN). To compute that we need to know TN.

Example. Suppose 1000 people were screened, with 985 true negatives, 5 false negatives and 10 positives, and of those 10 positives only 4 were true positives. Then the false positive rate would be 6/(6+985) = 0.605%.

That names for some of the various things you can compute from FP, TP, FN, and TN are given here [1].

With medical screening you are often have the case that only a small fraction of the people being screened have the condition. Even if a test is wrong only a small fraction of the time, (large fraction of people without the condition) x (small chance of them getting a false positive) can easily be as big as (small fraction of people with the condition) x (large chance of getting a true positive).

Continuing with the example from above, what screening does is take you from "We've got 1000 people and probably around 10 have the condition" to "here's 10 people and probably around 4 have the condition". You can then use more expensive tests that wouldn't be feasible to do on 1000 people on those 10 to find the ones who really have it.

[1] https://en.wikipedia.org/wiki/Confusion_matrix

OK, you're right, and I have done this before. My bad.

However, the inverse but highly relevant questions the patient asks are:

Given that I got a positive result, what's the chance that I actually have cancer?

Given that I got a negative result, what's the chance that I actually have it anyway?

I could attempt this, but right now I have something else to do. Will check back later and probably find that someone's answered those!

If I'm not interested in watching 23andMe flail trying to sell personal health info, does anyone have a recommendation for private whole-genome sequencing? I found Nebula at $550. Is there anywhere else people can recommend where I keep ownership?

They already completely burned any trust I had in them. If they split into two companies, that will just be two companies that I wouldn't trust with anything.

Is ElevenAndYourLawyer registered?

So that company is on the verge of bankruptcy but their CEO got a 32$M payout last year?

As the prophesies foretold, doom shall not impede a healthy executive bonus.

So the DNA company is becoming RNA?

Either that or it becomes CH11

Meiosis or Mitosis?

Reply on news.ycombinator.com