3 ms·

"They started with 13,700 people from the eMERGE Network, and looked for associations between 135,000 Neanderthal genetic variants and 1,689 different traits."

"They started with 13,700 people from the eMERGE Network, and looked for associations between 135,000 Neanderthal genetic variants and 1,689 different traits." Does anyone know how they compensated the statistics to control for the p-hacking?

adenadel 11y ago
I took a brief look at the paper scanning for multiple hypothesis testing corrections. "Neandertal variants have been hypothesized to influence many phenotypes in AMHs, including lipid metabolism, immunity, depression, digestion, hair, and skin, on the basis of the enrichment of Neandertal variants in regions of the genome relevant to these traits. Accordingly, we first tested these hypotheses using genome-wide complex trait analysis (GCTA) to estimate the phenotypic risk explained by 1495 genotyped common (minor allele frequency > 1%) Neandertal SNPs for a set of 46 highprevalence phenotypes from the hypothesized categories, using age, sex, and eMERGE site as covariates." When testing these 46 traits they controlled the FDR < 0.05 (false discovery rate). Then they replicated the associations in an independent sample. The effect sizes are pretty small (1-2%), so having the explanatory variants won't have much bearing whether or not you'll get these complex traits.
- nerdponx 11y ago
  Was the only correlation that survived FDR control and the validation set? I'm also curious as to why ethnic origin wasn't a covariate.
Houshalter 11y ago
You missed the next sentence: "They then checked any links they found against a second group of 14,700 eMERGE volunteers." So if any correlations were found, they basically did another study to replicate it. And that's a big sample size too. Many studies are published with far weaker correlations.
- azakai 11y ago
  That's true, but the effect size here is also tiny - 1%. It's plausible it's statistically significant but completely meaningless. It's possible this study is novel mainly for the methodology, which as reported is the first wide-scale test of its nature. The media of course focus on the correlations because they are more sensational - it gets people thinking "am I depressed because I'm part neanderthal?"
  - Houshalter 11y ago
    Well the base rate of being depressed is low, so an increase of 1% is kind of significant. If the base rate is only 1%, then a 1% increase means you are twice as likely to be depressed! I agree it's meaningless, in the sense that we have no idea what the cause is. It could be indirect, it could be random rather than anything special about neanderthals, it could be confounded with ethnicities that have higher neanderthal DNA, etc.
    - azakai 11y ago
      I think it's a 1% explanatory effect of the total variance. That means that 1% of the variance in the population (not the size of the effect) is explainable by this factor. (Also, the base rate of depression is far more than 1%, see e.g. http://tech.mit.edu/V120/N3/dep1.3n.html http://tech.mit.edu/V120/N3/dep1.3n.html - it's around 17% for just a major depressive episode.)

adenadel 11y ago

I took a brief look at the paper scanning for multiple hypothesis testing corrections. "Neandertal variants have been hypothesized to influence many phenotypes in AMHs, including lipid metabolism, immunity, depression, digestion, hair, and skin, on the basis of the enrichment of Neandertal variants in regions of the genome relevant to these traits. Accordingly, we first tested these hypotheses using genome-wide complex trait analysis (GCTA) to estimate the phenotypic risk explained by 1495 genotyped common (minor allele frequency > 1%) Neandertal SNPs for a set of 46 highprevalence phenotypes from the hypothesized categories, using age, sex, and eMERGE site as covariates." When testing these 46 traits they controlled the FDR < 0.05 (false discovery rate). Then they replicated the associations in an independent sample. The effect sizes are pretty small (1-2%), so having the explanatory variants won't have much bearing whether or not you'll get these complex traits.

nerdponx 11y ago
Was the only correlation that survived FDR control and the validation set? I'm also curious as to why ethnic origin wasn't a covariate.

Houshalter 11y ago

You missed the next sentence: "They then checked any links they found against a second group of 14,700 eMERGE volunteers." So if any correlations were found, they basically did another study to replicate it. And that's a big sample size too. Many studies are published with far weaker correlations.

azakai 11y ago
That's true, but the effect size here is also tiny - 1%. It's plausible it's statistically significant but completely meaningless. It's possible this study is novel mainly for the methodology, which as reported is the first wide-scale test of its nature. The media of course focus on the correlations because they are more sensational - it gets people thinking "am I depressed because I'm part neanderthal?"
- Houshalter 11y ago
  Well the base rate of being depressed is low, so an increase of 1% is kind of significant. If the base rate is only 1%, then a 1% increase means you are twice as likely to be depressed! I agree it's meaningless, in the sense that we have no idea what the cause is. It could be indirect, it could be random rather than anything special about neanderthals, it could be confounded with ethnicities that have higher neanderthal DNA, etc.
  - azakai 11y ago
    I think it's a 1% explanatory effect of the total variance. That means that 1% of the variance in the population (not the size of the effect) is explainable by this factor. (Also, the base rate of depression is far more than 1%, see e.g. http://tech.mit.edu/V120/N3/dep1.3n.html http://tech.mit.edu/V120/N3/dep1.3n.html - it's around 17% for just a major depressive episode.)