Comprehensive Assessment and Benchmark of Deep Generative Models for Proteolysis TArgeting Chimera (PROTAC) Design.

Proteolysis TArgeting Chimera (PROTAC) has been constantly proven to be an effective strategy for targeted protein degradation. More recently, various deep generative models have been proposed and applied in all lead compound discovery tasks, including PROTAC design. However, no quantitative assessment for the performance of these deep generative models on the PROTAC design task has yet been conducted. In this study, we provided a comprehensive overview of different kinds of the latest deep generative models including PROTAC-specific design models and general linker design models that could be utilized in the PROTAC design task. Then, representative structures from both classes were quantitatively evaluated on a benchmark of 40 experimental protein-ligand complex structures, together with a general de novo design model serving as an ablated model. This work aims to discuss the features and the generative performance of different types of molecular generative models for the PROTAC design task and help researchers to better apply these models in practical cases.