What are the best approaches (metrics, or a combination of metrics) to identify keystone taxa in microbial co-occurrence networks?
A valid way is finding the taxa that present the highest betweenness centrality, which means the taxa that have the highest interconnection between paths, which are not necessarily the taxa that present the highest number of connections. Ecologically, those are the taxa that are supposed to hold the network stability, also called hub or keystone taxa. In Gephi, for example, you can easily visualize this parameter in the node table.
As far as I have read, there is no consensus on what is the best method to infer Keystone taxa.
Here is our opinion about it (see the section "How to identify network hubs and what is their relevance for keystone taxa inference?")
https://www.researchgate.net/publication/352809565_Anti-tick_microbiota_vaccines_how_can_this_actually_work
Hey Dasiel,
The inference of a "keystone" depends on the question one is asking. Typically, most people go with taxa with the highest degree, whilst others chose another centrality measure such as Betweenness, as Andressa suggested. There are several centrality measures to choose from aside from, others being eigen etc. Here's a nice explanation of what other measures can be: Article Application of Graph Theory for Identifying Connectivity Pat...
We use a combination of two or more centralities in our lab, and then verify these findings either via literature or follow-up experiments in the lab. It is not impossible to design synthetic communities, expensive yes, but still doable. If one can validate these findings with additional omics strategies assessing their cooperation or exclusivity via gene networks, then it further adds weight to one's findings.
Having said that, it seems to me that you asked the question and answered it yourself with a publication from your own group. Maybe you wanted to raise your profile points ;). Either way, there is no one way to identify keystones without experimental validation.
Dear Susheel,
I appreciate your answer, I agree that the selection of the metrics depends on the research question, however, there is a big difference between the inferences made according to different metrics (i.e, betweenness centrality Vs Eigen centrality) in microbial co-occurrence networks, do you have an example from your lab on this?
No, the reason for the question here was honest, at that time (Jun 2020) I was reviewing the subject, I was looking for the opinion of the forum on this. A year later I have posted our opinion, with examples that we have been evaluating (inference plus confirmation). Our sincere intention is to have feedbacks and continue to improve our approaches. I hope to have other opinions, examples, etc. from you and other colleagues.
Best wishes,
Dasiel
Hello Dasiel, I've come across this same question in my last manuscript. You are right, there is no consensus regarding how to determine which taxa are the keystone. I suggest that the choice should be backed by the literature of the researcher area or a correlated one. In my case (Soil Science) I used both Banerjee et al.,2018 (https://doi.org/10.1038/s41396-019-0383-2) and Berry and Widder, 2014 (https://doi.org/10.3389/fmicb.2014.00219) to determine Keystone ASVs as nodes with degree ≥ 50, closeness centrality ≥ 0.25, and betweenness centrality ≤ 0.75. I exported the node/edge table and parsed it in R.
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biochemical Techniques
- Analytical Chemistry
- pH