Deng8 is not about multimorbidity #1
Description
This value in MOKGConfiguration/TABLE/MOKG/deng8.yaml is thought provoking:
triple_object:
value_for_encoding: multimorbidity
The authors don’t actually claim that they are associating proteins with “multimorbidity”. Rather, they are associating proteins with a particular cluster’s “multimorbidity level”. These are within-cluster associations, so the association p-values of a protein with different clusters’ “multimorbidity level” may vary. Indeed, each protein has potentially many different rows in the table, each with its own p-value (which one can see by sorting the table on protein names). The biological meaning in a Translator context of a cluster’s “multimorbidity level” is not real clear. If we consider a particular cluster (e.g., Cluster 1 as illuminated in main-text panel 4B), this cluster appears to be what I might call “liver diseases”. In that context, BST2 is a “multimorbidity” associated protein. But that means (I think, if I follow the authors’ logic) that BST2 is associated with lots of liver diseases. And it does not necessarily follow that it is associated with lots of diseases. But it could follow. If A is associated with lots of Bs. And Bs are a subclass of Cs. Then it would seem to follow that A is associated with lots of Cs. However, what would our propositional calculist say if the number of Bs was << the number of Cs? Or if a clinical doctor’s definition of multimorbidity was weighted towards having multiple diseases that were largely distinct from each other (which is sort of the case)? Technically, I do not think this is so, but if you want to get at nuances, try asking your favorite AI:
Is a clinical doctor’s definition of multimorbidity weighted towards having multiple diseases that are largely distinct from each other? For example, is having stroke and hypertension together less a degree of multimorbidity than having cancer and dementia? (edited)
2 replies
Thanks for highlighting that. It seems to me that this may be yet another case of ‘how to refer to a cluster resulting from an analysis’ (and likely other interpretation/modeling issues).
As for the last paragraph in italics, Jenn and I had multiple conversations last year about how multimorbidity is ‘counted’. If I recall correctly, she said most existing/prevalent methods simply count diseases. I was a proponent of adjusting weights…

Jared Roach
(https://hoodlabisb.slack.com/archives/C07B5B6G15J/p1758052915099509?thread_ts=1758052005.245229&cid=C07B5B6G15J)
Most definitions of multimorbidity require the diseases to be chronic, as well. The authors of Deng et al. did not filter their diseases on “chronic”. My overall gut sense is that this table is not primetime for Translator.
One hates to ignore potentially useful information. And clearly there is a lot here that perhaps some brilliant insight could leverage. But in 2025, if I was parsing information from this paper for Translator, I would go after the more atomic relationships, such as all the computed protein-disease relationships.
“Furthermore, 107,158 significant protein-incident disease associations were observed at a Bonferroni-corrected threshold of p < 2.59 × 10−8 (p < 0.05/[2,920∗660]) (Figure 1C).”
One (BIG) caveat: they don’t appear to have provided this information. (edited)