The goal of de novo design is to identify novel active compounds that can simultaneously satisfy a combination of essential optimization goals such as activity, selectivity, physical-chemical and ADMET properties. It is a process of generating novel compounds from scratch that are normally not present in databases and are not being previously considered in the context of the given target. Identifying a compound that meets all these criteria is still far from trivial.
In this talk we will discuss how AI has been adopted for the needs of de novo design of small molecules. We will illustrate this in the context of REINVENT, which is a platform developed internally within AstraZeneca and made open source. It is based on using different flavors of generative models that know how to generate small molecules. The generative models are be placed in various learning scenarios and can learn to generate molecules that satisfy a diverse set of criteria thus elegantly serving the needs of de novo design.