Fig. 1From: Applying atomistic neural networks to bias conformer ensembles towards bioactive-like conformationsSchematic representation of AtNNs architecture. AÂ AtNNs take as input a conformation in the form of atomic numbers with corresponding atomic coordinates to output a single value for the conformation. BÂ Architecture of an interaction block. A new atom embedding is produced by using an input atom embedding and basis functions (encoding the 3D structure) through a message passing paradigm performing convolutions between atoms in a neighbourhood subgraph defined by a distance cut-offBack to article page