Stuctural Analysis of Molecular Networks: AMES Mutagenicity

The characterization of chemical compounds based on their molecular graphs is an important task for identifying properties such as toxicity or mutagenicity. We used different groups of topological descriptors using the AMES mutagenicity data. Instead of optimizing the classification performance, the aim of this study is to perform a structural analysis of the underlying set of molecular graphs to gain better insights of the data set. The structural analysis identifies two groups of molecular networks. One group contains graphs with linear patterns (outliers), and the other group contains graphs that exhibit patterns of regular graphs (remainders). We show that the set of used topological descriptors chosen for this study cannot capture enough group-specific structural information within the remainders group to achieve the discrimination ability of the outliers group. Finally, this leads us to the conclusion that it is necessary to identify existing or develop new descriptors that capturing specific structural information to achieve better discrimination ability.