Fig. 11
From: Comparing structural fingerprints using a literature-based similarity benchmark
Directed graphs summarising the relative performance on a the single-assay benchmark, b the multi-assay benchmark and c the Riniker–Landrum benchmark. All fingerprints higher in the graph are better than those below if a path between them exists. The numbers indicate the net difference (number of times better minus number of times worse) for neighbouring fingerprints. Note that the scales are different as the upper values are bounded by the dataset size: 3629 for a, 4563 for b, and 88 for c. Note that the net differences may not obey the triangle inequality. To find out the net difference for a non-neighbouring pair, see the tables in the Additional file 1