Given a set of chemical compounds, chemical data mining is to characterize the compounds present in the data set and apply a variety of mining methods to discover relationships between the compounds and their biological and chemical activities.
In 1969, Hansch [6] introduced quantitative structure-activity relationship (QSAR) analysis which attempts to correlate physicochemical or structural properties of compounds with biological and chemical activities. These physicochemical and structural properties are determined empirically or by computational methods. QSAR prefers vectorial mappings of compounds, which are usually coded by existing physicochemical and structural fingerprints. Dehaspe et al. [3] applied inductive logic programming to predict chemical carcinogenicity by mining frequent substructures in chemical datasets, which identifies new structural fingerprints so that QSAR could build comprehensive analytical models.
