Hacker News Reader

Comment by hprotagonist | Hacker News Reader

In general, i regard GWAS as fishing expeditions that are more or less guaranteed to find spurious correlations. Von Neumann's old saw about "With four parameters I can fit an elephant, and with five I can make him wiggle his trunk." absolutely applies here without other information.

That said, finding associations with ACE2 and blood type regions are not totally insane!

View on HN

My money is on the virus being an aerosol and badly ventilated indoor events accounting for most of the transmissions. There is growing evidence for this and also that the amount of virus particles you get into your system determines how sick you get (related to your overall health).

If this is true, social distancing wouldn't help us much but disallowing gatherings of people in badly ventilated areas would.

https://www.nytimes.com/2020/06/02/opinion/coronavirus-super...

https://medicalxpress.com/news/2020-06-coronavirus-importanc...

https://youtu.be/SnI5ZsvvTrg

> There is growing evidence for this and also that the amount of virus particles you get into your system determines how sick you get (related to your overall health).

Which would be a good argument for variolation.

That is, deliberately infect yourself with a very small virus dose. Small enough that you don't get seriously ill, but big enough that you get immunity.

To you and the parent, this is why I think we see large antibody counts in prison and military populations while not have as many cases. Folks were infected with a very small initial viral load and their immune systems fought it off and developed antibodies.

This is also why we see young healthy medical personnel die from the disease, 1) their immune system is degraded from harsh working conditions 2) the viral loads are many multiples higher from basically getting coughed in the face all day long

I’m interested: why would a prison or military population fit with a very small viral load?

Dense populations with small cross sections, either at their stations or in bunks/cells. Mostly aerosolized transmission vs direct contact.

If this were 1800, sure. You could also just skip the phase 3 trials on the Sinovac vaccine instead, but it is not a well supported argument.

I know variolation is banned by 2020 regulations in most of the world.

That doesn't mean it's a bad idea.

With a large enough sample size that starts to go away. We're starting to see real benefits from GWAS, it just took 1.5 decades to get the sample sizes up to a level where the data generated was worth looking at.

My thoughts exactly. The article seemed to suggest that they looked at a truly ginormous number of SNP's, which suggests a high susceptibility to the xkcd "green M&M" type problem (https://xkcd.com/882/).

However, the good thing about blood type is that it should be reasonably easy/quick to do a followup where you examine a lot more patients, or even just randomly selected individuals, to see if it really holds up. As a type O person myself, I am hopeful.

A GWAS will use significance criteria that correct for the millions of multiple comparisons. (This paper does.)

GWAS studies typically use a p-value cutoff of 5*10^-8 to avoid exactly that problem.

p values cutoff can't salvage that

as a matter of fact artificially raising the cutoff to make the results sound more convincing is a form a p-hacking

it is exactly the opposite of how p-values work. a smaller p-value does not make a result more reliable, it makes no sense whatsoever to reject a 10^-7 but trust in 10^-8

the misuse of p-values in the world is rampant - the only people that don't misuse them are those that do not use them ;-)

See here: https://en.wikipedia.org/wiki/Genome-wide_significance

"The most commonly accepted threshold is p < 5 × 10−8, which is based on performing a Bonferroni correction for all the independent common SNPs across the human genome"

The reason for the low cutoff is to correct for multiple hypothesis testing.

Any what do you mean by "opposite of how p-values work"?

Obviously if P(data or more extreme|null hypothesis) = 10^-8 it's significantly less likely to see the data if the null hypothesis was true, than if P(data or more extreme| null hypothesis) = 10^-7.

They used the "Global Screening Array (GSA), version 2.0 (Illumina), which contains 712,189 variants" [0]

[0] https://www.nejm.org/doi/full/10.1056/NEJMoa2020283?query=fe... "Genomewide Association Study of Severe Covid-19 with Respiratory Failure"

Doesn’t plaqunial work by altering Ace2?

Reply on news.ycombinator.com