Abstract
Geometric deep learning (GDL), which is based on neural network architecturesthat incorporate and process symmetry information, has emerged as a recentparadigm in artificial intelligence. GDL bears particular promise in molecularmodeling applications, in which various molecular representations withdifferent symmetry properties and levels of abstraction exist. This reviewprovides a structured and harmonized overview of molecular GDL, highlightingits applications in drug discovery, chemical synthesis prediction, and quantumchemistry. Emphasis is placed on the relevance of the learned molecularfeatures and their complementarity to well-established molecular descriptors.This review provides an overview of current challenges and opportunities, andpresents a forecast of the future of GDL for molecular sciences.