REINVENT 4: A Modern AI Molecular Design Platform
Hannes H Loeffler, Jiazhen He, Alessandro Tibo, Jon Paul Janet, Alexey Voronov, Lewis Mervin and Ola Engkvist,
MolecularAI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden
We introduce the next release of REINVENT which will be version 4, a major update in functionality from the previous release. REINVENT is a modern open source generative model platform for the design of small molecules and is based on recurrent neural networks (RNN) and, increasingly, transformer architectures.
Various design methods are seamlessly supported within general machine learning frameworks like transfer learning (TL), reinforcement learning (RL) and curriculum learning (CL). REINVENT enables and facilitates de novo design, R-group replacement and library design, linker search and scaffold hopping, functionality for molecular optimization as well as synthesis scoring and prediction.
