Bayesian statistics improves biological interpretability of metabolomics data from human cohorts, 2023, Brydges, Che, Lipkin and Fiehn

Preprint Abstract

Background: Univariate analyses of metabolomics data currently follow a frequentist approach, using p-values to reject a null-hypothesis. However, the usability of p-values is plagued by many misconceptions and inherent pitfalls. We here propose the use of Bayesian statistics to quantify evidence supporting different hypotheses and discriminate between the null hypothesis versus lack of statistical power.

Methods: We use metabolomics data from three independent human cohorts that studied plasma signatures of subjects with myalgic encephalomyelitis / chronic fatigue syndrome (ME/CFS). Data are publicly available, covering 84-197 subjects in each study with 562-888 identified metabolites of which 777 were common between two studies, and 93 compounds reported in all three studies. By comparing results from classic multiple regression against Bayesian multiple regression we show how Bayesian statistics incorporates results from one study as prior information into the next study, thereby improving the overall assessment of the likelihood of finding specific differences between plasma metabolite levels and disease outcomes in ME/CFS.

Results: Whereas using classic statistics and Benjamini-Hochberg FDR-corrections, study 1 detected 18 metabolic differences, study 2 detected no differences. Using Bayesian statistics on the same data, we found a high likelihood that 97 compounds were altered in concentration in study 2, after using the results of study 1 as prior distributions. These findings included lower levels of peroxisome-produced ether-lipids, higher levels of long chain, unsaturated triacylglycerides, and the presence of exposome compounds that are explained by difference in diet and medication between healthy subjects and ME/CFS patients. Although study 3 reported only 92 reported compounds in common with the other two studies, these major differences were confirmed. We also found that prostaglandin F2alpha, a lipid mediator of physiological relevance, was significantly reduced in ME/CFS patients across all three studies.

Conclusions: The use of Bayesian statistics led to biological conclusions from metabolomic data that were not found through the frequentist analytical approaches more commonly employed. We propose that Bayesian statistics to be highly useful for studies with similar research designs if similar metabolomic assays are used.

https://www.biorxiv.org/content/10.1101/2022.05.17.492312v1?ct=

Edit: Published Sept 2023, see post #4

 

I wonder if the results would change if they changed the order of the studies.

Here they started with the Nagy-Szakal et al. study then used its results as a prior for the study by Che et al. And then used those results as a prior for the Naviaux et al. study. But if you reversed or changed this order, would you find different results?

 

maybe the answer can be found at https://www.sciencedirect.com/topics/mathematics/bayes-rule ?

disclaimer: i am completely incapable of anything.

 

Now published, open access, https://www.mdpi.com/2218-1989/13/9/984

 

Did they look at / adjust for when in any female patients’ menstrual cycle samples were taken?

(ETA: this might be a dumb question. I’m only asking because I recall that prostaglandin is related (some kind of feedback thing?) to progesterone which oscillates through the menstrual cycle - causal relationship. So, if someone were having menstrual cycles that would affect things and if someone were not having menstrual cycles low levels might be normal. But I don’t know anything, just wondered)