Fig. 2
From: Comprehensive evaluation of pure and hybrid collaborative filtering in drug repurposing
(a) Benchmarking training and testing pipeline iterated \(N=100\) times for drug repurposing for a specific algorithm, a splitting method for training/testing and validation subsets, and a validation metric. Note that the training/testing subsets are always split at random. (b) Correlogram of metrics collected during the benchmark on randomly split training and testing sets, referring to metrics in Table 3. The total number of considered values is then \(N=18,700\) (see Table 9 in Appendix). The lower triangle of the plot shows linear regressions between each pair of metrics, with the corresponding \(R^2\) when greater than 0.25. The upper triangle displays the Spearman’s \(\rho\) correlations between each pair of metrics. The diagonal shows the empirical frequency distribution of values for each metric.
