Reproduce “3DLinker: An E(3) Equivariant Variational Autoencoder for Molecular Linker Design” 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

Yinan Huang, Xingang Peng, Jianzhu Ma, Muhan Zhang

3DLinker presents an E(3) equivariant variational autoencoder designed to jointly generate 2D molecular graphs and accurate 3D coordinates for molecular linkers. The model achieves an RMSD of 0.079 Å for 3D coordinate accuracy and a 93.58% graph recovery rate, outperforming previous methods while also predicting optimal anchor atom locations on input fragments.

Key insights

  • Jointly generating 2D molecular graphs and accurate 3D coordinates within an E(3) equivariant framework is essential for producing geometrically feasible and chemically valid linkers.
  • Directly predicting optimal anchor atoms on molecular fragments within the generation process removes a significant bottleneck in practical linker design.
  • An iterative global coordinate update strategy during sequential generation is crucial for achieving high recovery rates and ensuring overall structural integrity.
  • E(3) equivariant latent representations capture meaningful chemical and spatial information, proving useful for downstream property prediction tasks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.