NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation
Zhiyuan Liu, Yanchen Luo, Han Huang, Enzhi Zhang, Sihang Li, Junfeng Fang, Yaorui Shi, Xiang Wang, Kenji Kawaguchi, Tat-Seng Chua
2025-02-20
Summary
This paper talks about NExT-Mol, a new AI model that combines two different approaches to generate 3D molecules more effectively. It's like creating a super-smart chemistry assistant that can design new molecules in 3D, which is really important for developing new drugs and materials.
What's the problem?
Creating 3D molecules using AI is tricky because you need to balance two things: making sure the molecules are valid (meaning they could actually exist) and making them in 3D shapes. Previous methods were good at one or the other, but not both. It's like trying to draw a realistic 3D picture of an animal that doesn't exist in nature - you need to make sure it looks 3D and also looks like a real animal.
What's the solution?
The researchers created NExT-Mol, which uses two steps to generate 3D molecules. First, it uses a language model (like a really smart autocomplete for molecule structures) to create valid 2D molecule designs. Then, it uses another AI model to turn these 2D designs into 3D shapes. They made this work better by making the language model bigger, improving the 3D model, and teaching the 3D model some tricks from the 2D model.
Why it matters?
This matters because it could speed up the process of discovering new drugs and materials. By generating more accurate and diverse 3D molecules, scientists can explore new possibilities faster and more efficiently. It's like giving chemists a super-powered imagination tool that can come up with new molecule ideas that are more likely to work in the real world. This could lead to breakthroughs in medicine and materials science, potentially helping to create new treatments for diseases or develop advanced materials for technology.
Abstract
3D molecule generation is crucial for drug discovery and material design. While prior efforts focus on 3D diffusion models for their benefits in modeling continuous 3D conformers, they overlook the advantages of 1D SELFIES-based Language Models (LMs), which can generate 100% valid molecules and leverage the billion-scale 1D molecule datasets. To combine these advantages for 3D molecule generation, we propose a foundation model -- NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation. NExT-Mol uses an extensively pretrained molecule LM for 1D molecule generation, and subsequently predicts the generated molecule's 3D conformers with a 3D diffusion model. We enhance NExT-Mol's performance by scaling up the LM's model size, refining the diffusion neural architecture, and applying 1D to 3D transfer learning. Notably, our 1D molecule LM significantly outperforms baselines in distributional similarity while ensuring validity, and our 3D diffusion model achieves leading performances in conformer prediction. Given these improvements in 1D and 3D modeling, NExT-Mol achieves a 26% relative improvement in 3D FCD for de novo 3D generation on GEOM-DRUGS, and a 13% average relative gain for conditional 3D generation on QM9-2014. Our codes and pretrained checkpoints are available at https://github.com/acharkq/NExT-Mol.