A new graph modelisation for molecule similarity

TL;DR

This paper introduces a novel graph representation focusing on molecular cycles to improve the efficiency and relevance of molecule similarity comparisons, aiding retrosynthesis analysis.

Contribution

It proposes a new cycle-based graph model for molecules and an algorithm to extract these graphs, enhancing similarity measurement methods.

Findings

Cycle-based graphs reduce computational complexity.

Graphs of cycles improve similarity relevance.

Method demonstrates effectiveness on selected molecules.

Abstract

In order to define the process of restrosynthesis of a new organic molecule, it is often necessary to be able to draw inspiration from that of a molecule similar to the target one of which we know such a process. To compute such a similarity, an oftently used approach is to solve a Maximum Common Edge Subgraph (MCES) problem on molecular graphs, but such an approach is limited by computation time and pertinence of similarity measurement. In this paper, we define and analyse here a new graph representation of molecules to algorithmically compare them. The purpose is to model the structure of molecule by a graph smaller than the molecular graph and representing the interconnexion of its elementary cycles. We provide an algorithm to efficiently obtain such a graph of cycles from a molecular graph. Then by solving MCES problems on those graphs, we evaluate the pertinence of using graphs of cycles for molecular similarity on a select set of molecules.