[Paper Review] Diffusion Models for Molecules: A Survey of Methods and Tasks

Liang Wang, Chao Song+5 more|ArXiv.org|Feb 13, 2025

Diffusion Coefficients in Liquids5 citations

TL;DR

This paper surveys diffusion-model-based molecular generation, organizing the field by diffusion formulations, molecular data modalities, and generated tasks, and provides a taxonomy and future directions.

ABSTRACT

Generative tasks about molecules, including but not limited to molecule generation, are crucial for drug discovery and material design, and have consistently attracted significant attention. In recent years, diffusion models have emerged as an impressive class of deep generative models, sparking extensive research and leading to numerous studies on their application to molecular generative tasks. Despite the proliferation of related work, there remains a notable lack of up-to-date and systematic surveys in this area. Particularly, due to the diversity of diffusion model formulations, molecular data modalities, and generative task types, the research landscape is challenging to navigate, hindering understanding and limiting the area's growth. To address this, this paper conducts a comprehensive survey of diffusion model-based molecular generative methods. We systematically review the research from the perspectives of methodological formulations, data modalities, and task types, offering a novel taxonomy. This survey aims to facilitate understanding and further flourishing development in this area. The relevant papers are summarized at: https://github.com/AzureLeon1/awesome-molecular-diffusion-models.

Motivation & Objective

• Provide an up-to-date overview of diffusion-model-based molecular generative methods.
• Introduce a taxonomy categorizing work by method formulations, data modalities, and task types.
• Summarize key diffusion-model formulations and their molecular applications.
• Identify gaps, challenges, and future directions to guide further research.

Proposed method

• Review core diffusion-model formulations: DDPMs, score-based models (SMLD), and stochastic differential equations (SDEs).
• Explain discrete (D3PM) and latent (LDM) variants and how they apply to molecular data.
• Categorize molecular data modalities into 2D/topological space, 3D geometric space, and joint spaces.
• Map diffusion formulations to molecular generative tasks (de novo, optimization, conformer generation, docking, etc.).
• Provide a systematic taxonomy (method formulations, data modalities, task types) to organize the literature.

Experimental results

Research questions

• RQ1 What are the main diffusion-model formulations used for molecular generation and their differences?
• RQ2 How do molecular data modalities (2D, 3D, joint) influence diffusion-model design and performance?
• RQ3 What molecular generative tasks are addressed by diffusion models, and how are they categorized?
• RQ4 What are the current gaps and future directions in diffusion-model-based molecular design?

Key findings

• An up-to-date, systematic overview of diffusion-model-based molecular generation.
• A novel taxonomy organizing work by method formulations, data modalities, and task types.
• A synthesis of representative models and tasks across 2D, 3D, and joint spaces.
• Identification of opportunities and challenges to guide future research in diffusion-based molecular design.
• Provision of a public resource listing related works at the project GitHub page.