ProxelGen: Generating Proteins as 3D Densities

2025-06-24

AI Summary

This work addresses the limited geometric flexibility of conventional point-cloud representations in protein structure generation by proposing a novel 3D voxelized density (proxel)-based representation. Methodologically, we introduce a generative framework integrating a 3D CNN variational autoencoder (VAE) with a latent-space diffusion model, marking the first application of continuous density-field modeling to proteins—enabling fine-grained, shape-conditioned control. Our contributions are threefold: (1) pioneering density-based representation replaces discrete point clouds, substantially enhancing geometric plasticity and conditional controllability; (2) on motif-scaffold design benchmarks, our method achieves superior novelty and Fréchet Inception Distance (FID) scores compared to current state-of-the-art models; and (3) it maintains high designability, yielding physically plausible structures with functional potential. This paradigm establishes a new foundation for conditional, de novo protein design.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Humans and AI: Game Design — Procedural Content Generation & Storytelling

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

We develop ProxelGen, a protein structure generative model that operates on 3D densities as opposed to the prevailing 3D point cloud representations. Representing proteins as voxelized densities, or proxels, enables new tasks and conditioning capabilities. We generate proteins encoded as proxels via a 3D CNN-based VAE in conjunction with a diffusion model operating on its latent space. Compared to state-of-the-art models, ProxelGen's samples achieve higher novelty, better FID scores, and the same level of designability as the training set. ProxelGen's advantages are demonstrated in a standard motif scaffolding benchmark, and we show how 3D density-based generation allows for more flexible shape conditioning.

Problem

• Generating protein structures as 3D densities
• Enabling new tasks via voxelized density representation
• Improving novelty and designability in protein generation

Innovation

• Generates proteins as 3D voxelized densities
• Uses 3D CNN-based VAE with diffusion model
• Enables flexible shape conditioning capabilities