Table 2.

List of evaluation approaches.

Title	Evaluation	Graph	Correct predictions	Methods
Evaluation 1	Remove random 30% of links from the DDIs (repeat 10 times)	Comprehensive evaluation for recovery of DDS similarity links	Match predicted links with the actual ones	All methods are applied
Evaluation 2	Remove random 30% of links (repeat 10 times)	Ground Truth using DrugBank	Match predicted links with DrugBank reported interactions	SP_2 (the best from evaluation 1 over disjoint graph) and RA (the best from evaluation 1 over joint graph)
Evaluation 3	Remove random 30% of links (repeat 10 times)	Whole graph including DDS, FDS, FFS	Match predicted links with the actual ones	SP_2 (the best from evaluation 1 over disjoint graph) and RA (the best from evaluation 1 over joint graph)
Evaluation 4	Prepare a list of gold standard food-drug interactions extracted from the literature. These interactions will be hidden from any training and will be used to measure and evaluate the validity of FDMine	Whole graph including DDS, FDS, FFS	Match predicted links with the actual food-drug interactions (gold standard dataset)	SP_2 (the best from evaluation 1 over disjoint graph) including SP_3 and RA (the best from evaluation 1 over joint graph) including AA, CN, and L3