HyperDiffusionFields (HyDiF):Diffusion-Guided Hypernetworks for Learning Implicit Molecular Neural Fields
CME Department Seminar
August 6, 2025
11:00 AM - 12:00 PM America/Chicago
Location
Room 1047 ERF and Zoom
Address
842 W. Taylor St., Chicago, IL 60607
Calendar
Download iCal FilePresenter: Sudarshan Babu, PhD, Chan Zuckerberg Biohub Chicago
Location: Room 1047 ERF and Zoom
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, termed Molecular Neural Fields (MNFs). To enable learning across molecules and facilitate generalization, we use a shared hypernetwork that, conditioned on a molecule, generates the weights of its MNF. For generative capabilities, we train the hypernetwork as a denoising diffusion model, enabling sampling in the function space of molecular fields. Our design naturally supports masked diffusion, allowing structure-conditioned generation tasks such as molecular inpainting by selectively noising regions of the field. Beyond generation, the localized and continuous nature of MDFs allows for spatially fine-grained feature extraction for molecular property prediction—an advantage over graph- or point cloud-based methods. Finally, we show that HyDiF scales to larger biomolecules, highlighting its potential for field-based molecular modeling.
Speaker Bio: Sudarshan Babu is an AI fellow at the Chan Zuckerberg Biohub Chicago, advised by Aly Khan. His research expertise is in deep learning with a PhD in computer science from TTIC. He has published multiple papers at venues like ICML, NeurIPS, and CVPR, and works at the Chan Zuckerberg Institute. His research focuses on generative models, molecular modeling, and 3D representation learning.
Date posted
Jul 30, 2025
Date updated
Jul 30, 2025