Total: 1
This work introduces FMG, a field-based model for drug-like molecule generation. We show how the flexibility of this method provides crucial advantages over the prevalent, point-cloud based methods, and achieves competitive molecular stability generation. We tackle optical isomerism (enantiomers), a previously omitted molecular property that is crucial for drug safety and effectiveness, and thus account for all molecular geometry aspects. We demonstrate how previous methods are invariant to a group of transformations that includes enantiomer pairs, leading them invariant to the molecular R and S configurations, while our field-based generative model captures this property.