ProxelGen: Generating Proteins as 3D Densities

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Faltings, Felix, Stark, Hannes, Barzilay, Regina, Jaakkola, Tommi
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908419768713216
author Faltings, Felix
Stark, Hannes
Barzilay, Regina
Jaakkola, Tommi
author_facet Faltings, Felix
Stark, Hannes
Barzilay, Regina
Jaakkola, Tommi
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_19820
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ProxelGen: Generating Proteins as 3D Densities
Faltings, Felix
Stark, Hannes
Barzilay, Regina
Jaakkola, Tommi
Biomolecules
Machine Learning
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.
title ProxelGen: Generating Proteins as 3D Densities
topic Biomolecules
Machine Learning
url https://arxiv.org/abs/2506.19820