Fig. 2From: Task-driven knowledge graph filtering improves prioritizing drugs for repurposingPredicted probabilities of positive and negative relations on the original dataset (A, C) and the modified dataset (B, D) of Hetionet (A, B) and DRKG (C, D). Shown are 200.000 positive and 200.000 negative relations from the training set, scored and batch normalized in batches of 2000 each. The predictions are obtained from the best performing ComplEx modelsBack to article page