Machine learning prediction of the potential pesticide applicability of three dihydroquinoline derivatives: syntheses, crystal structures and physical properties

Abstract

Increasingly machine learning processes have been applied in the search and development of compounds that may have specific physicochemical properties to the desired application. This article describes how a machine learning model led us to the synthesis of three dihydroquinoline derivatives with potential application as a pesticide. The synthesized compounds were predicted to be active against the Tobacco mosaic virus (≅ 90%) and Fusarium oxysporum (≅ 78%). Regarding a correlation between the pesticide activity and the molecular structure, the new dihydroquinoline derivatives were structurally characterized using spectroscopic techniques and single crystal X-ray diffraction. They crystallized into orthorhombic (I) and monoclinic (II and III) crystal systems with supramolecular arrangements maintained primarily by non-classical C–H⋯O hydrogen bonds, which form dimers and chains in their molecular packaging. Frontier molecular orbitals and molecular electrostatic potential maps were undertaken using density functional theory in order to study the electronic properties of the observed molecular conformations. Finally, the developed approach is a useful tool on new pesticide investigation when experimental toxicity data are not available.