Fig. 4

Gene-feature network. The gene-feature network (A) with its modules enriched with the Gene Ontology biological pathway terms (B–D) and causal networks composed of 29 GIMs based on conditional (in)dependency augmented with principles of Mendelian randomization (E–F). A Gene-feature network is constructed using MAGMA results (false discovery rate < 0.01). One-to-one associations are found for 67 gene-feature pairs; however, the other genes and features are interconnected to form modules. B A module with four features and the genes enriched with dADP (deoxyadenosine diphosphate) biosynthetic process. C Twelve genes associated with a feature (m/z 467.256, RT 278.1) are over-represented for nucleosome assembly. D Another feature (m/z 522.734, RT 43.2) is associated with the seven genes functioning plasma membrane long-chain fatty acid transport. E, F Two modules identified from the causal network analysis of 29 GIMs. Annotated metabolites in each module provide information about the unannotated metabolites in the same module specially if they are highly connected (e.g., Ne,Ne dimethyllysine (m/z 175.144, RT 104.5) in (E). In the module in (F), a strong dependency (p = 5 × 10^–13) between an unannotated feature (m/z 188.9574, RT 44.2) and fenson (m/z 269.0023, RT 125.5) may provide annotation information for the feature