Applications of modular co-design for de novo 3D molecule generation

Abstract

De novo 3D molecule generation is a pivotal task in drug discovery. However, many recent geometric generative models struggle to produce high-quality geometries, even if they able to generate valid molecular graphs. To tackle this issue and enhance the learning of effective molecular generation dynamics, we present Megalodon – a family of scalable transformer models. These models are enhanced with basic equivariant layers and trained using a joint continuous and discrete denoising co-design objective. We assess Megalodon's performance on established molecule generation benchmarks and introduce new 3D structure benchmarks that evaluate a model's capability to generate realistic molecular structures, particularly focusing on geometry precision. We show that Megalodon achieves state-of-the-art results in 3D molecule generation, conditional structure generation, and structure energy benchmarks using diffusion and flow matching. Furthermore, we demonstrate that scaling Megalodon produces up to 49× more valid molecules at large sizes and 2–10× lower energy compared to the prior best generative models. The code and the model are available at https://github.com/NVIDIA-Digital-Bio/megalodon.