Molecular Generative Model via Retrosynthetically Prepared Chemical Building Block Assembly

Abstract

Deep generative models are attracting attention as a smart molecular design strategy. However, previous models often render molecules with low synthesizability, hindering their real-world applications. Here, a novel graph-based conditional generative model which makes molecules by tailoring retrosynthetically prepared chemical building blocks until achieving target properties in an auto-regressive fashion is proposed. This strategy improves the synthesizability and property control of the resulting molecules and also helps learn how to select appropriate building blocks and bind them together to achieve target properties. By applying a negative sampling method to the selection process of building blocks, this model overcame a critical limitation of previous fragment-based models, which can only use molecules from the training set during generation. As a result, the model works equally well with unseen building blocks without sacrificing computational efficiency. It is demonstrated that the model can generate potential inhibitors with high docking scores against the 3CL protease of SARS-COV-2.

GRELinker: A Graph-Based Generative Model for Molecular Linker Design with Reinforcement and Curriculum Learning

“…Another test case for linker generation is to optimize the linker within the protein-binding site, aiming to increase the binding affinity. Usually, the docking score is used as a surrogate for binding affinity with the target protein, and it has been widely employed as the objective for conditional molecular generation in de novo drug design . Here, we integrated the docking score in the reward function of RL to fine-tune the prior model to generate molecules with improved binding affinity.…”

Gotta be SAFE: a new framework for molecular design

“…To effectively apply generative approaches in live drug discovery projects, it is essential to incorporate goal-directed generation, guiding generation of novel molecules towards specic properties. Therefore, we follow established methodologies 37,38 to assess the model's ability for goal-directed generation using simple molecular properties. More precisely, we optimize toward specic values for key molecular properties, including Topological Polar Surface Area (TPSA), Molecular Weight (MW), Calculated LogP (CLOGP), and Quantitative Estimation of Druglikeness (QED).…”

Deep Generative Models in De Novo Drug Molecule Generation

“…By using these motifs as building blocks in the generation process, the accuracy of the generation process can be effectively improved, especially for large molecules such as polymers . The main methods for building motif vocabulary are (1) defined by functional groups; (2) constructed based on heuristic rules; (3) using retrosynthetically prepared chemical building blocks; (4) mining connection-aware motifs . Motif-level building blocks can also be used in conjunction with atom-level building blocks to better utilize the information in the motifs …”