Finding Largest Common Substructures of Molecules in Quadratic Time

Abstract

Finding the common structural features of two molecules is a fundamental task in cheminformatics. Most drugs are small molecules, which can naturally be interpreted as graphs. Hence, the task is formalized as maximum common subgraph problem. Albeit the vast majority of molecules yields outerplanar graphs this problem remains $\mathsf{NP}$-hard.

We consider a variation of the problem of high practical relevance, where the rings of molecules must not be broken, i.e., the block and bridge structure of the input graphs must be retained by the common subgraph. We present an algorithm for finding a maximum common connected induced subgraph of two given outerplanar graphs subject to this constraint. Our approach runs in time $\mathcal {O}(\varDelta n^2)$ in outerplanar graphs on n vertices with maximum degree $\varDelta$. This leads to a quadratic time complexity in molecular graphs, which have bounded degree. The experimental comparison on synthetic and real-world datasets shows that our approach is highly efficient in practice and outperforms comparable state-of-the-art algorithms.

This work was supported by the German Research Foundation (DFG), priority programme “Algorithms for Big Data” (SPP 1736).