Fig 3. Intersection & union policy effect on indication ranking performance.

Positive (approved drug-disease indications) and negative (non-indications) ensemble prediction distributions were visualized. a: Intersection policy effect on ensemble ranking distribution of positive and negative sets. b: Union policy effect on ensemble ranking distribution of positive and negative sets. X-axis represents ensemble combinations between CBR, probCBR, Rephetio, TransE, DistMult, ComplEx and RotatE of sizes two through seven. Boxplot distributions highlight the first, second and third quartiles for Positive and Negative prediction ranks. The distribution between Positive and Negative sets are statistically significantly different regardless of algorithm combination length and ensemble policy. Violin plots highlight the kernel density estimate for each policy’s Positive and Negative rank distribution; the y-axis is log scaled.