Fig. 3
From: Predicting clinically promising therapeutic hypotheses using tensor factorization
Benchmark performance of leave one out experiments. Model performance on predicting clinical outcomes of target classes (a) and disease clusters (b) in the leave-one-out experiments in terms of Area Under Receiving Operation Curve (AUROC, x-axis) and Area Under Precision Recall Curve (AUPRC, y-axis). 95% confidence interval is calculated using 1000 bootstraps. Dotted lines mark the AUROC (vertical) and AUPRC (horizontal) of a random guess, which is 0.5 and the fraction of positives in the testing set, respectively. The percentage of target-indication pairs in each held-out set is listed after the pipe symbol (|) in the titles. (LR: Logistic Regression; GBM: Gradient Boosting Machine; RF: Random Forest; MF: Matrix Factorization; BTF: Bayesian Tensor Factorization)