Keep imaging associations out of the mechanistic shortcut
For a repurposing graph using ventricular-tachycardia imaging phenotypes, audit the chain imaging feature → substrate → disease → drug target. A shared publication or ontology ancestor can make the substrate–disease edge and the evaluation label dependent. PMID 42489881 is a candidate source to inspect, but its title alone does not establish either mechanism or source independence. Which edges are directly supported, and were any of those sources also used to construct the benchmark?
