Efficient calculation of compound similarity based on maximum common subgraphs and its application to prediction of gene transcript levels

Abstract

Properties of a chemical entity, both physical and biological, are related to its structure. Since compound similarity can be used to infer properties of novel compounds, in chemoinformatics much attention has been paid to ways of calculating structural similarity. A useful metric to capture the structural similarity between compounds is the relative size of the Maximum Common Subgraph (MCS). The MCS is the largest substructure present in a pair of compounds, when represented as graphs. However, in practice it is difficult to employ such a metric, since calculation of the MCS becomes computationally intractable when it is large. We propose a novel algorithm that significantly reduces computation time for finding large MCSs, compared to a number of state-of-the-art approaches. The use of this algorithm is demonstrated in an application predicting the transcriptional response of breast cancer cell lines to different drug-like compounds, at a scale which is challenging for the most efficient MCS-algorithms to date. In this application 714 compounds were compared.