Skip to main content


Fig. 6 | Journal of Cheminformatics

Fig. 6

From: A probabilistic molecular fingerprint for big data settings

Fig. 6

Comparing measured distances between MHFP6 and ECFP4 (2048-D) in different data sets. The distances in all data sets show moderate to strong linear correlation of r = 0.659, r = 0.792, and r = 0.829 for GDB-13 hydrocarbons, Drugbank, and MMP respectively. a, d While the distribution of measured distances is similar for MHFP6 and ECFP4 for the GDB-13 subset, ECFP4 seems to measure a distance of 0.0 between clearly different molecules. Gaps appear in measured ECFP4 distances, resulting in a multimodal distribution. b, e The distances measured in Drugbank show a strong correlation. Both fingerprints seem to measure a distance of 1.0 in molecules where a finer grained distance measure could proof beneficial. c, f The MMP data set exposes the inability of ECFP4 to distinguish between highly similar molecules that differ only in the size of one ring whereas MHFP6 seems to express higher resolution for distance measurements between very similar compounds

Back to article page