Non a Priori Automatic Discovery of 3D Chemical Patterns: Application to Mutagenicity

This article introduces a new type of structural fragment called a geometrical pattern . Such geometrical patterns are defined as molecular graphs that include a labelling of atoms together with constraints on interatomic distances. The discovery of geometrical patterns in a chemical dataset relies on the induction of multiple decision trees combined in random forests. Each computational step corresponds to a refinement of a preceding set of constraints, extending a previous geometrical pattern . This paper focuses on the mutagenicity of chemicals via the definition of structural alerts in relation with these geometrical patterns . It follows an experimental assessment of the main geometrical patterns to show how they can efficiently originate the definition of a chemical feature related to a chemical function or a chemical property. Geometrical patterns have provided a valuable and innovative approach to bring new pieces of information for discovering and assessing structural characteristics in relation to a particular biological phenotype.