De novo Molecular generation based on GCPN and GraphAF

Abstract

There is increasing adoption of artificial intelligence in drug discovery. De novo molecular design involves the creation of novel molecules with desiredproperties and functions. The process is often computationally intensive andrequires the exploration of vast chemical space. In recent years, deep learning-based approaches have emerged as powerful tools for De Novo molecular design. These methods involve the use of deep neural networks to learn patterns in molecular structures and properties, enabling the generation of new molecules with desired properties. One popular approach is the use of generative mod- els, To De Novo , we construct two dataset to date, namely MUV and BACE such as Graph Convolutional Policy Network (GCPN) and GraphAF, which learn to generate molecules that mimic a given dataset. We will implement two graph generative models GCPN and GraphAF. We first pretrain both models on MUV and BACE datasets. We finetune both models with reinforcement learning to optimize two properties of generated molecule , After conducting a comprehensive analysis of each model, we have determined that GCPN outperforms GraphAF, delivering significantly better results