De Novo Drug Design

The computational design of novel drug-like molecules from scratch, either by assembling molecular fragments or by generating structures using generative models, without starting from a known active compound. Modern de novo design uses deep generative models (VAEs, GANs, reinforcement learning) to propose synthetically accessible molecules optimized for multiple objectives simultaneously.

示例

• REINVENT — generative model using recurrent neural networks with RL
• Fragment-based growing using FBDD and computational linking
• Generative adversarial network (GENTRL) designing DDR1 kinase inhibitors