Fig. 6

Predictive explanations generated by querying matching rules on the KG This figure showcases the example of the digenic gene pair MYH7-ANKRD1, part of the independent test set and predicted as disease-causing with the highest probability. A Subgraph extracted by traversing all paths (excluding those traversing “Phenotype”, “Disease” and “OligogenicCombination” nodes) of a length \(\le\) 3. A total of 342 paths, 127 nodes and 447 edges exists. B Top 5 matching rules ranked by their associated probability score. Each rule is written in their abbreviated form (see Table 1) with its conditions separated by &. Indices for node types (e.g. BP\(_1\)) are used in unification conditions (e.g. BP\(_1\)=BP\(_2\)) to constrain entities to be the same across different metapaths. The numerical value associated with each metapath (e.g. \(\ge\) 0.21) sets the path reliability threshold, which conditions the minimum path reliability score of all underlying paths. We display the number of paths obtained by querying the KG with the rule with that specific gene pair. C Returned explanation subgraph for the 1st rule based on the 7 matching paths. D Returned explanation subgraph for the 2nd rule based on the 5 matching paths. Entity types are represented with the same colors as in (A). Explanations subgraphs for the 3rd, 4th and 5th rules are provided in Additional file 1: Figs. I