Skip to main content
. 2023 Aug 21;12:giad057. doi: 10.1093/gigascience/giad057

Table 3:

The performance comparison of MOA prediction between KGML-xDTD and different baseline models (e.g., MultiHop and KGML-xDTD w/o DP) based on the test set (described in “Data split” section). The metrics in this table are calculated using path scores and all non-DrugMechDB-matched 3-hop paths between drug and disease as “negative” paths for each true-positive drug–disease pair (see more details in “MOA prediction evaluation method” section)

Model MPR MRR Hit@1 Hit@10 Hit@50 Hit@100 Hit@500
MultiHop 61.400% 0.027 0.017 0.042 0.067 0.118 0.345
KGML-xDTD w/o DP 72.965% 0.015 0.008 0.017 0.067 0.160 0.403
KGML-xDTD (ours) 94.696% 0.109 0.059 0.193 0.496 0.613 0.849

The abbreviation “w/o DP” in the name of model “KKGML-xDTD w/o DP” represents “without using demonstration paths.”