Fig. 5: Comparative analysis of prediction patterns across knowledge graph embedding (KGE), pretrained, and combined models.
From: A fused deep learning approach to transform drug repositioning
a Probability density distribution of prediction scores showing distinct distributional characteristics. b Violin plots quantifying prediction score statistics with median and mean values. c Percentile analysis of score distribution reveals model-specific ranking patterns. d–f Correlation scatter plots demonstrating pairwise relationships between KGE-Pretrained, KGE Combined, and Pretrained-Combined models with Pearson correlation coefficients. g Overlap percentage analysis of top 10% predictions between model pairs. h Comparison of high-frequency drug recommendations (frequency ≥10) across models. i Venn diagram showing overlap of high-frequency drug recommendations between the three prediction models.
