Evaluation of knowledge graph embedding approaches for drug-drug interaction prediction in realistic settings

BMC Bioinformatics

Table 2 Evaluation of different embedding methods in various CV schemes

		Traditional CV			Drug-wise CV			Pairwise CV			Time-slice CV
Embedding	ML Model	AUPR	F1	AUC	AUPR	F1	AUC	AUPR	F1	AUC	AUPR	F1	AUC
RDF2Vec	Logistic Regression	0.78	0.71	0.78	0.76	0.69	0.76	0.73	0.66	0.74	0.75	0.68	0.76
CBOW	Naive Bayes	0.68	0.63	0.70	0.68	0.63	0.70	0.68	0.63	0.70	0.71	0.67	0.73
	Random Forest	0.92	0.85	0.92	0.79	0.69	0.78	0.75	0.64	0.74	0.80	0.69	0.80
RDF2Vec	Logistic Regression	0.79	0.72	0.79	0.77	0.70	0.77	0.75	0.68	0.75	0.76	0.69	0.76
SG	Naive Bayes	0.76	0.68	0.74	0.75	0.68	0.74	0.75	0.67	0.73	0.78	0.72	0.78
	Random Forest	0.92	0.85	0.93	0.81	0.71	0.80	0.76	0.63	0.75	0.80	0.68	0.80
TransE	Logistic Regression	0.78	0.70	0.76	0.73	0.67	0.73	0.72	0.67	0.72	0.75	0.68	0.76
	Naive Bayes	0.75	0.69	0.73	0.72	0.68	0.71	0.72	0.68	0.71	0.76	0.72	0.76
	Random Forest	0.90	0.83	0.91	0.76	0.69	0.77	0.73	0.64	0.73	0.77	0.65	0.78
TransD	Logistic Regression	0.74	0.68	0.74	0.74	0.67	0.74	0.72	0.66	0.72	0.74	0.70	0.75
	Naive Bayes	0.72	0.68	0.71	0.72	0.67	0.71	0.72	0.67	0.71	0.73	0.70	0.73
	Random Forest	0.91	0.84	0.91	0.77	0.69	0.77	0.73	0.64	0.73	0.78	0.68	0.78

Bio2RDF DrugBank knowledge graph and DDIs from DrugBank v5 were used in the evaluation. We considered these CV settings: traditional CV, disjoint CV (drug-wise, pairwise) and time-slice CV. The settings are explained in the Evaluation section. (Bold: best score)

ISSN: 1471-2105