Straight-Line Diffusion Model for Efficient 3D Molecular Generation

Wei-Ying Ma (Tsinghua University) · Yanyan Lan (Institute for AI Industry Research (AIR), Tsinghua University) · Shikun Feng (Tsinghua University, Tsinghua University) · Yuyan Ni (Tsinghua University; Chinese Academy of Sciences) · Haohan Chi (Tsinghua University) · Bowen Zheng (Huazhong University of Science and Technology) · Huan-ang Gao (Tsinghua University) · Zhi-Ming Ma (Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Chinese Academy of Sciences)
3d molecule generationbenchmarksdiffusion-based modelsenhancement techniquesgenerative processlearning efficiencylinear trajectorymolecular generationmolecular structuresnoise sensitivityreconstruction effortsampling efficiencysampling stepsstate-of-the-art performancestraight-line diffusion model

Diffusion-based models have shown great promise in molecular generation but often require a large number of sampling steps to generate valid samples. In this paper, we introduce a novel Straight-Line Diffusion Model (SLDM) to tackle this problem, by formulating the diffusion process to follow a linear trajectory. The proposed process aligns well with the noise sensitivity characteristic of molecular structures and uniformly distributes reconstruction effort across the generative process, thus enhancing learning efficiency and efficacy. Consequently, SLDM achieves state-of-the-art performance on 3D molecule generation benchmarks, delivering a 100-fold improvement in sampling efficiency.