Josh Rackers Research - 3D molecule generation by denoising voxel

Journal article

Pedro H. O. Pinheiro, Joshua A. Rackers, J. Kleinhenz, Michael Maser, O. Mahmood, Andrew Watkins, Stephen Ra, Vishnu Sresht, Saeed Saremi
arXiv.org, 2023

Semantic Scholar ArXiv DBLP DOI

Cite

APA Click to copy
Pinheiro, P. H. O., Rackers, J. A., Kleinhenz, J., Maser, M., Mahmood, O., Watkins, A., … Saremi, S. (2023). 3D molecule generation by denoising voxel grids. ArXiv.org.

Chicago/Turabian Click to copy
Pinheiro, Pedro H. O., Joshua A. Rackers, J. Kleinhenz, Michael Maser, O. Mahmood, Andrew Watkins, Stephen Ra, Vishnu Sresht, and Saeed Saremi. “3D Molecule Generation by Denoising Voxel Grids.” arXiv.org (2023).

MLA Click to copy
Pinheiro, Pedro H. O., et al. “3D Molecule Generation by Denoising Voxel Grids.” ArXiv.org, 2023.

BibTeX Click to copy

@article{pedro2023a,
  title = {3D molecule generation by denoising voxel grids},
  year = {2023},
  journal = {arXiv.org},
  author = {Pinheiro, Pedro H. O. and Rackers, Joshua A. and Kleinhenz, J. and Maser, Michael and Mahmood, O. and Watkins, Andrew and Ra, Stephen and Sresht, Vishnu and Saremi, Saeed}
}

Abstract

We propose a new score-based approach to generate 3D molecules represented as atomic densities on regular grids. First, we train a denoising neural network that learns to map from a smooth distribution of noisy molecules to the distribution of real molecules. Then, we follow the neural empirical Bayes framework [Saremi and Hyvarinen, 2019] and generate molecules in two steps: (i) sample noisy density grids from a smooth distribution via underdamped Langevin Markov chain Monte Carlo, and (ii) recover the ``clean'' molecule by denoising the noisy grid with a single step. Our method, VoxMol, generates molecules in a fundamentally different way than the current state of the art (i.e., diffusion models applied to atom point clouds). It differs in terms of the data representation, the noise model, the network architecture and the generative modeling algorithm. VoxMol achieves comparable results to state of the art on unconditional 3D molecule generation while being simpler to train and faster to generate molecules.