Fig. 6.

Bayesian network estimates involving systolic blood pressure (BP), mean diffusivity of the external capsule WM tract (EC) and three covariates (Age, Income, grey matter (GM) volume). Estimated edges are denoted with dashed arrows. (A) Standard BN estimate using the hill-climbing algorithm with BIC in bnlearn. Five SNPs (with known associations with the connected phenotype) are used as genetic “anchors”, i.e., the known associations are provided as prior information and are represented as fixed edges in the graph (indicated as solid arrows). Edges between SNPs and those that would have Age or any SNP as effect (endpoint of an arrow) were excluded a priori. (B) Ancestral graph estimate based on the RFCI algorithm in the pcalg R package; no prior information on presence or absence of individual edges was provided. No edges were found for two SNPs. Dotted, bi-directional arrows indicate the presence of a common cause. An edge without arrowheads means that directionality of the relationship could not be determined from the data (e.g., Age–Income). A $★$ next to an edge indicates the potential presence of a latent, unmeasured variable.