If I’m looking at the association between cardiovascular disease (CVD) risk and gut microbial beta diversity, I understand that this would show the association between levels CVD risk and inter-individual differences in gut microbiome composition. However, I’m finding understanding that association in a biological context a bit tricky. For example, if it is a positive significant association, is that suggesting that harbouring a more distinct gut microbial community compared to somebody else , may be linked with a CVD risk (i.e there is no “core” microbiome signature associated with CVD risk per se that would be common across individuals, more likely to be different compositions when comparing).
I'm more familiar with comparing beta diversity between disease vs control, rather than association with an endpoint, so any explanations would be appreciated - thank you.
I think the correct interpretation would be "CVD risk is a significant factor that can explain a certain amount of variance in the gut microbial composition."
In fact, no matter it is an endpoint variable or a disease vs control scenario, this kind of association just tests whether a certain variable is a significant factor affecting the microbiome composition/structure. In the disease vs control case, the variable under test is just "disease status".
Man Kit Cheung
Thank you for your answer! Please could you let me know whether that intepretation would still hold if you have CVD risk as a dependent variable and beta diversity as an independent variable? For example, I am using the first three principal components from a principal co-ordinate analysis (from the original beta diversity dissimilarity matrix) as individual indepndent variables in a multivariate linear regression model.
I have also looked at this using a Permanova using 999 permutations (using the CVD risk scores and the n x n dissimilarity matrix).
Ok, so you are actually doing it the other way around; in your regression analysis, you are testing whether gut microbiome beta diversity is a significant explanatory variable affecting CVD risk. A positive significant association would mean that a more different
gut microbial composition is associated with/leads to/predicts a higher CVD risk.
I may be wrong but it seems that beta (and alpha) diversity is more commonly set as the dependent variable in these kinds of analysis.
Man Kit Cheung Thank you very much again for your reply!
So perhaps that might suggest that there is no core signature per se that can predict CVD risk, but understand I would need to run other analyses to support this.
Yes, you're right, beta diversity is usually a dependent variable in a model.
I think I have seen one paper (Article Microbial Diversity and Abundance of Parabacteroides Mediate...
where they examined the association between blood pressure as a dependent variable and the first three principal co-ordinates from a Bray-Curtis dissimilarity matrix.As a last question, please could you advise whether it would make more sense to use principal components from a PCA to capture the variation in the relative abundance of the microbial community (genera level) as a whole, and then examine the association with CVD risk as a dependent variable? I guess what I would like to find out is if there is any association with CVD risk at the "community level", rather than considering individual genera or alpha diversity. I'm not entirely sure the best way to do this.
Thanks again
Yes, you can use the PCoA axes to calculate the association between CVD risk and the overall microbial community structure.
But I would recommend calculating the associations between CVD risk and alpha diversity and individual genera/taxa as well, unless you have strong reasons not to do so. They are just testing different things and may give you interesting findings.
Man Kit Cheung
Thank you! Yes, I've looked at the association between CVD risk and alpha diveristy/genera taxa.
As a final question, please could you let me know for the PCoA - which distance metric you would use? For example, if using PCA, would you use the Aitchison distance or would you stick with calculating a bray-curtis dissimilarity matrix and generate principal co-ordinates from this?
Thanks again!
My personal preference would still be PCoA based on arguably the most commonly used Bray-Curtis and UniFrac distances although I also acknowledge that theoretically Aitchison distance, which takes into account the compositionality nature of microbiome data, should be kind of superior to these metrics.
Again, these metrics have different emphasis/strengths and it is not a bad idea to compute more than one of them, especially in case the one you chosen initially doesn't give you good/expected results.
Man Kit Cheung
Thank you very much for all of your help! I've selected the answer in line with the initial question.
Thanks again
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Nucleic Acids
- RNA