4 ms·

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 "gre

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/ 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.

pjscott 6y ago
A GWAS will use significance criteria that correct for the millions of multiple comparisons. (This paper does.)
bobcostas55 6y ago
GWAS studies typically use a p-value cutoff of 5*10^-8 to avoid exactly that problem.
- glofish 6y ago
  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 ;-)
  - drocer88 6y ago
    See here: https://en.wikipedia.org/wiki/Genome-wide_significance 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"
  - zosima 6y ago
    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.
drocer88 6y ago
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=featured_home https://www.nejm.org/doi/full/10.1056/NEJMoa2020283?query=fe... "Genomewide Association Study of Severe Covid-19 with Respiratory Failure"

pjscott 6y ago

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

bobcostas55 6y ago

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

glofish 6y ago
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 ;-)
- drocer88 6y ago
  See here: https://en.wikipedia.org/wiki/Genome-wide_significance 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"
- zosima 6y ago
  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.

drocer88 6y ago

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