Similarity Measures: Is It Possible To Compare Dissimilar Structures?

The determination of the similarity between dissimilar compounds is a problem only partially solved. The necessity of defining unconnected description spaces to evaluate the diversity of chemical structures is an important drawback of many existing methodologies. Our objective is thus the definition of a method capable of making acceptable comparisons between any kind of structures. The definition of a virtual similarity index, indirectly calculated from standard similarity analyses, and the selection of a proper representation space allow the calculation of two new similarity indexes: the global and the local similarity indexes. The performance of the method is discussed by its application to three sets of compounds at different complexity levels.