Fig. 10
From: Comparing structural fingerprints using a literature-based similarity benchmark
Histograms showing the relative performance of a subset of the fingerprints relative to the HashTT fingerprint in the Riniker–Landrum benchmark. Relative performance was measured for each of the 50 repetitions by counting the number of times a particular fingerprint had a higher BEDROC(20) value compared to the HashTT fingerprint and subtracting the number of times it had a lower value. The difference between the performance of the ECFC4 and HashTT fingerprints was not found to be statistically significant. All others fingerprints shown were either better (those to the right) or worse (those to the left) than HashTT. A bin width of 5 was used