[tinokid]

C'mon people...this is not science, this is fortune telling. They are just doing a bunch of huge multiparameter curve fits, which are going to pick up signal from all sorts of things that have nothing to do with biology no matter how big their sample size is. These people will tell you with a straight face that a bunch of markers that just happen to be associated with "Northeast European ancestry" are also predictive of "Polka dancing ability," and their papers are shuffling and repackaging thousands upon thousands of nonsensical little tidbits exactly like that. Life experience doesn't just "average out." Do better.

[ggggtez]

Do you believe statistics aren't real math? I'm not sure why the complaint. Sure it may be inaccurate, but as others said, along questions like: "do you have X heritage?" are already used as predictors for certain diseases. Genetic investigations just gets a little closer to the base truth, even if no one knows exactly which genes are doing which things.

[Balgair]

I think GP is talking about Type I errors; where you go hunting for correlations that are p<0.05 in a data set.

https://en.wikipedia.org/wiki/Type_I_and_type_II_errors

[tinokid]

But genetics may or may not be the "base truth" in regard to disease risk. It's difficult to tease out the effects of genes as opposed to environment. When your model has thousands or millions of genetic loci this only gets harder.

[Ultimatt]

Some GWAS studies thats definitely true, but ultimately if the only signal you're picking up from the genetics is someones ancestry (haplotype) and that is enough to confer accurate predictions then Im not sure what your problem is? Thats great, plenty of people in the US for example have no understanding about what their genetics might confer because they might not even know where their genetic heritage comes from.

[laretluval]

Have you read the paper and determined they are not taking adequate precautions against overfitting?