Benchmarking and Validation of JChem ECFP and FCFP Fingerprints
The cornerstone of pharmaceutical chemistry is Crum Brown’s observation that similar compounds have similar therapeutic benefits. Cheminformatics tries to capture this insight by defining measures of similarity between the computer representations of two molecules, with the hope of capturing a medicinal chemist’s intuitive sense of “likeness”, and thereby correlate with bioactivity. This poster evaluates the chemical similarity measures offered by ChemAxon on the standard reference Briem and Lessel benchmark. Any such benchmark must by necessity be flawed; the similarity between two molecules is influenced by the framework by which they are compared. However a robust similarity measure should typically perform better on such benchmarks, whilst a weaker model of chemical similarity would be expected to perform worse (on average). The results show that ChemAxon’s chemical fingerprints perform as well as or better than traditional cartridge fingerprint methods, whilst the new in 5.4 ECFP fingerprints equal the best chemical similarity methods published to date.