Finding Characteristic Substructures for Metabolite Classes

Abstract

We introduce a method for finding a characteristic substructure for a set of molecular structures. Different from common approaches, such as computing the maximum common subgraph, the resulting substructure does not have to be contained in its exact form in all input molecules. Our approach is part of the identification pipeline for unknown metabolites using fragmentation trees. Searching databases using fragmentation tree alignment results in hit lists containing compounds with large structural similarity to the unknown metabolite. The characteristic substructure of the molecules in the hit list may be a key structural element of the unknown compound and might be used as starting point for structure elucidation. We evaluate our method on different data sets and find that it retrieves essential substructures if the input lists are not too heterogeneous. We apply our method to predict structural elements for five unknown samples from Icelandic poppy.