Straight-Line Diffusion Model for Efficient 3D Molecular Generation

2025-03-04

AI Summary

Diffusion models for 3D molecular generation suffer from excessive sampling steps and low efficiency. To address this, we propose Linear Trajectory Diffusion (LTD), a novel paradigm that replaces conventional nonlinear noise scheduling with a geometry-aware linear schedule. In LTD, noise addition and removal follow straight-line trajectories in the joint space of atomic coordinates and types, enabling uniform distribution of reconstruction load and alignment with structural noise sensitivity. Methodologically, we design a lightweight denoising network coupled with a joint coordinate–type representation mechanism. On mainstream 3D molecular generation benchmarks, LTD achieves state-of-the-art performance while reducing sampling steps by 100×, maintaining high validity and chemical reasonableness. We further validate its cross-modal generalizability on synthetic (toy) data and image generation tasks. To our knowledge, this is the first work to introduce linear trajectory modeling into diffusion-based generative frameworks, establishing a new paradigm for efficient geometric generation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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. Furthermore, experiments on toy data and image generation tasks validate the generality and robustness of SLDM, showcasing its potential across diverse generative modeling domains.

Problem

• Improves sampling efficiency in 3D molecular generation.
• Reduces required sampling steps for valid molecular samples.
• Enhances learning efficiency and efficacy in generative models.

Innovation

• Introduces Straight-Line Diffusion Model (SLDM)
• Linear trajectory enhances sampling efficiency
• Achieves 100-fold improvement in sampling efficiency