Reproduce “LEGO-MOF: Equivariant Latent Manipulation for Editable, Generative, and Optimizable MOF 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
Chaoran Zhang, Guangyao Li, Dongxu Ji
The LEGO-MOF framework introduces a continuous, SE(3)-equivariant latent space for Metal-Organic Framework (MOF) linkers, enabling fine-grained editing, zero-shot isoreticular expansion, and property-driven optimization of MOFs. It demonstrates the ability to generate chemically valid linkers and improve CO2 uptake in 74.0% of optimized MOFs, achieving a mean relative boost of 147.5% in pure CO2 uptake.
Key insights
- A continuous, disentangled, and SE(3)-equivariant latent representation for MOF linkers enables fine-grained, differentiable manipulation while preserving chemical and topological integrity, especially through robust anchor atom preservation.
- Explicitly decoupling 3D spatial coordinates from invariant chemical features within the latent space allows for independent and precise control over MOF geometry and chemistry.
- Integrating a differentiable generative model with a high-fidelity surrogate property predictor facilitates gradient-based optimization of existing MOFs directly towards desired performance metrics, moving beyond predefined libraries.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.