Seems like there is a new study every week that shows just how influential guy microbes are. Like they literally control your mood, how you act, what illnesses you get or don’t get, how you look, pretty much everything about how you act on a daily basis.
Makes you wonder who is actually in charge of your body.
pumukl on
Well, I was having a coffee break and here’s what I noticed with this study:
Sex mismatch: the gut microbiota GWAS includes both men and women, but prostate cancer is exclusively male (the authors acknowledge that)
Problem with multiple testing: I think they tested here 473 bacterial taxa and identified 16 as significant at P-value < 0.05.
By chance alone, you’d expect around 24 false positives at this threshold. The reported p-values range from 0.002 to 0.048 -none would survive Bonferoni correction (with a threshold = 0.0001). There’s no mention of FDR correction for the initial screening.
I would say, that the causal claims are overstated given the methodological limitations. The machine learning validation is particularly unconvincing, and the multiple testing issue means many of the 16 identified taxa are likely false positives, I also could not find any heterogeneity statistics (Cochran Q results), leave-one-out sensitivity analyses to name a few.
In summary I would be very careful with taking these results into the clinic…
2 Comments
Seems like there is a new study every week that shows just how influential guy microbes are. Like they literally control your mood, how you act, what illnesses you get or don’t get, how you look, pretty much everything about how you act on a daily basis.
Makes you wonder who is actually in charge of your body.
Well, I was having a coffee break and here’s what I noticed with this study:
Sex mismatch: the gut microbiota GWAS includes both men and women, but prostate cancer is exclusively male (the authors acknowledge that)
Problem with multiple testing: I think they tested here 473 bacterial taxa and identified 16 as significant at P-value < 0.05.
By chance alone, you’d expect around 24 false positives at this threshold. The reported p-values range from 0.002 to 0.048 -none would survive Bonferoni correction (with a threshold = 0.0001). There’s no mention of FDR correction for the initial screening.
I would say, that the causal claims are overstated given the methodological limitations. The machine learning validation is particularly unconvincing, and the multiple testing issue means many of the 16 identified taxa are likely false positives, I also could not find any heterogeneity statistics (Cochran Q results), leave-one-out sensitivity analyses to name a few.
In summary I would be very careful with taking these results into the clinic…