Deep generative models for 3D molecular structure

Deep generative models have gained recent popularity for chemical design. Many of these models have historically operated in 2D space; however, more recently explicit 3D molecular generative models have become of interest, which are the topic of this article. Dozens of published models have been developed in the last few years to generate molecules directly in 3D, outputting both the atom types and coordinates, either in oneshot or adding atoms or fragments step-by-step. These 3D generative models can also be guided by structural information such as a binding pocket representation to successfully generate molecules with docking score ranges similar to known actives, but still showing lower computational efficiency and generation throughput than 1D/2D generative models and sometimes producing unrealistic conformations. We advocate for a unified benchmark of metrics to evaluate generation and propose perspectives to be addressed in next implementations.

Automated summary

Deep generative models have become popular in chemical design. This article focuses on explicit 3D molecular generative models, which have gained interest recently. Multiple models have been developed to generate molecules in 3D, providing atom types and coordinates. These models can be guided by structural information and produce molecules with similar docking scores to known actives, but they are less efficient and sometimes generate unrealistic conformations. The article advocates for a unified benchmark of metrics and proposes future perspectives to be addressed.