Reproduce “ProxelGen: Generating Proteins as 3D Densities” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Felix Faltings, Hannes Stark, Regina Barzilay, Tommi Jaakkola
ProxelGen introduces a protein generative model that operates on 3D density representations, termed proxels, instead of traditional atomistic coordinates. This approach enables fixed-dimension processing, allows for flexible spatial conditioning on arbitrary shapes, and generates diverse protein structures with competitive fidelity compared to existing atomistic models.
Key insights
- A 3D density representation, termed 'proxel', is a viable and advantageous alternative to atomistic point clouds for protein generative modeling.
- Using fixed-dimension voxel grids allows leveraging highly optimized 3D CNNs and latent diffusion models, offering new inductive biases for protein generation.
- Generating proteins as 3D densities enables flexible spatial conditioning, including arbitrary shape adherence and scaffolding of complex, disconnected motifs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.