HyperDiffusionFields (HyDiF): Diffusion-Guided Hypernetworks for Learning Implicit Molecular Neural Fields

2025-10-20

AI Summary

This work addresses the limitations of conventional discrete atomic-coordinate or graph-based representations in molecular generation, structural completion (e.g., repair), and fine-grained property prediction. We propose a novel continuous-field paradigm for 3D molecular conformation modeling, representing molecules as implicit neural fields guided by molecular directional fields. A conditional hypernetwork generates molecule-specific field functions, coupled with a denoising diffusion model to enable both unconditional generation and masked-conditional structural completion directly in function space. Key contributions include: (i) the first integration of implicit neural fields with diffusion models for molecular modeling; (ii) a diffusion-guided hypernetwork architecture enabling cross-molecule generalization; and (iii) scalable modeling—from organic small molecules to large biomolecules. Experiments demonstrate significant improvements over baselines in conformational sampling accuracy, structural completion robustness, and fine-grained prediction of intrinsic molecular properties.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We introduce HyperDiffusionFields (HyDiF), a framework that models 3D molecular conformers as continuous fields rather than discrete atomic coordinates or graphs. At the core of our approach is the Molecular Directional Field (MDF), a vector field that maps any point in space to the direction of the nearest atom of a particular type. We represent MDFs using molecule-specific neural implicit fields, which we call Molecular Neural Fields (MNFs). To enable learning across molecules and facilitate generalization, we adopt an approach where a shared hypernetwork, conditioned on a molecule, generates the weights of the given molecule's MNF. To endow the model with generative capabilities, we train the hypernetwork as a denoising diffusion model, enabling sampling in the function space of molecular fields. Our design naturally extends to a masked diffusion mechanism to support structure-conditioned generation tasks, such as molecular inpainting, by selectively noising regions of the field. Beyond generation, the localized and continuous nature of MDFs enables spatially fine-grained feature extraction for molecular property prediction, something not easily achievable with graph or point cloud based methods. Furthermore, we demonstrate that our approach scales to larger biomolecules, illustrating a promising direction for field-based molecular modeling.

Problem

• Models molecular structures as continuous 3D fields rather than discrete representations
• Enables generative molecular design through diffusion-based function space sampling
• Supports molecular property prediction via spatially fine-grained feature extraction

Innovation

• Hypernetworks generate molecular neural field weights
• Diffusion model enables function space sampling
• Masked diffusion supports structure-conditioned generation