express correlation between blood analytes and gut microbiome

[sierra-moxon]

We have a use case from Arpita J. (Multiomics Provider) to express correlations between blood analytes (or some other entity) and the gut microbiome. They have KEGG ortholog identifiers for the data representing the gut microbiome.

Perhaps we are trying to say "this KEGG function is correlated with the gut." This data is from the ISB Wellness data set.

It would be helpful to understand the kind of searches intended to bring back this data, so we can make sure we're modeling the critical bits that would be found via those searches. For example, is the "important" part of this statement that the KEGG function is correlated with the gut specifically, or that the KEGG function is correlated with a specific phenotype like "increased hemoglobin" (made up example phenotype), or even that the KEGG function is associated with the gut microbiome. (Or probably all three! :)).

tagging @aj95b (generator of this question) and @karthiksoman (SPOKE representative) for their feedback.

[realmarcin]

Hi Sierra — I'm involved in metagenome knowledge modeling with other hats so jumping in with 2 cents, early days still for the field! It looks like the statement is something like: “presence of KO function X in the gut microbiome (samples 1,2,3) correlates with change in blood metabolite Y” ?

For metagenomics these statements are almost always still correlative — but often that is not stated explicitly which generates a lot of confusion. A rare example of a more causal finding is the A20 locus and allergies https://science.sciencemag.org/content/349/6252/1106

[aj95b]

The entire ISB's wellness dataset is composed of samples from a cohort of largely healthy individuals, so as of now we are not looking at over-representation of a phenotype. We only want to present its (the gut microbiome's) statistical correlation with blood analytes in this cohort.

[karthiksoman]

Given the description of the data, what I understand is that wellness dataset gives statistical correlations between KO functions and blood analytes from a cohort of healthy individuals. In SPOKE, we have modelled microbiome pathways as nodes from kegg modules instead of KO. However, these pathway nodes do not have direct connection with Compounds (in this case analytes). Instead the path goes as: pathway--> EC--> Reaction --> Compounds. So, I think, one can check to trace this path in the knowledge network, observing the statistical correlation from the wellness dataset. It would be nice to bring Sergio on this discussion to get more ideas on the kind of searches possible on wellness dataset.

[nlharris]

Will discuss during Biolink Help Desk on Oct 11.

[sierra-moxon]

Notes from our biolink-model helpdesk call:

Notes from predicates WG meeting:

Action items:

• [x] add KEGG.ORTHOLOGY prefix to Molecular Activity
• [ ] add two associations: one between molecular activity and chemical entity, the other between MA and MA (if they don't already exist).

Also: automate additions of feature variables and statistical attributes via EPC modeling worksheet (with definitions) to biolink-model (three here, will come in a different ticket).