How to calculate the rule confidence of the Metapaths?

[LuMflowers]

Hi, could you share the code of calculating the rule confidence of the Metapaths? I didn't get the right rule confidence according to your desciption.

[liu-yushan]

Hi, do you mean the metapaths in Table 2? Since the confidence estimation is based on sampling, the confidences might vary slightly. Some relations are reflexive, e.g., CrC means the same as _CrC, and there are different formulations of one rule, e.g., ["CtD", "CrC", "CtD"] represents the same rule as ["CtD", "_CrC", "CtD"]. In such cases, we estimate the confidence for each representation separately and take the average. All equivalent formulations are also included in datasets/Hetionet/rules.txt.

[liu-yushan]

We also estimate the confidence of both the rule and the inverse of the rule, e.g., ["Compound", "CpD", "Disease"] and ["Disease", "_CpD", "Compound"] and take the average to get a better estimate.

[liu-yushan]

The script for calculating the confidences can be found here: https://github.com/liu-yushan/PoLo/tree/main/datasets/Hetionet/preprocessing.

[LuMflowers]

The script for calculating the confidences can be found here: https://github.com/liu-yushan/PoLo/tree/main/datasets/Hetionet/preprocessing.

Thank you for your code. But I still have a question about the files in ../datasets/Hetionet/preprocessing/, how can I get these files such as node_edges.json and metapath_p3.json.

[liu-yushan]

node_edges.json is a dictionary that lists all nodes' neighbors, grouped by relations and node types. You can just go through all triples and add the corresponding information to the dictionary. metapath_p3.json lists all possible metapaths between compounds and diseases up to length 3. This is based on the metagraph (see https://het.io/about/).