On the value of using 3D-shape and electrostatic similarities in deep generative methods

Multi-parameter optimization, the heart of drug design, is still an open challenge. Thus, improved methods for automated compounds design with multiple controlled properties are desired. Here, we present a significant extension to our previously described fragment-based reinforcement learning method (DeepFMPO) for the generation of novel molecules with optimal properties. As before, the generative process outputs optimized molecules similar to the input structures, now with the improved feature of replacing parts of these molecules with fragments of similar 3D-shape and electrostatics. By performing comparisons of 3D-fragments, we can simulate 3D properties while overcoming the notoriously difficult step of accurately describing bioactive conformations. The comparison of electrostatic potential and molecular shape is performed using the new ESP-Sim python package, allowing the calculation of state-of-the-art partial charges (e.g., RESP with B3LYP/6-31G**) obtained using the quantum chemistry program Psi4. The new improved method is demonstrated with a scaffold-hopping exercise identifying CDK2 bioisosteres. All code is open-source and freely available.