MOFDiff: Coarse-grained Diffusion for Metal-Organic Framework Design

Authors: Xiang Fu, Tian Xie, Andrew Scott Rosen, Tommi S. Jaakkola, Jake Allen Smith

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We comprehensively evaluate our model s capability to generate valid and novel MOF structures and its effectiveness in designing outstanding MOF materials for carbon capture applications with molecular simulations.
Researcher Affiliation Collaboration 1MIT CSAIL 2Microsoft Research AI4Science 3Department of Materials Science and Engineering, UC Berkeley 4Materials Science Division, Lawrence Berkeley National Laboratory
Pseudocode Yes Algorithm 1 Optimize building block orientations for MOF assembly
Open Source Code Yes Code available at https://github.com/microsoft/MOFDiff.
Open Datasets Yes We train and evaluate our method on the BW-DB dataset, which contains 304k MOFs with less than 20 building blocks (as defined by the metal-oxo decomposition algorithm) from the 324k MOFs in Boyd et al. 2019.
Dataset Splits Yes We use 289k MOFs (95%) for training and the rest for validation.
Hardware Specification No The paper does not specify any particular GPU or CPU models, or other specific hardware configurations used for running experiments.
Software Dependencies Yes MOFid-v1.1.0, MOFChecker-v0.9.5, egulp-v1.0.0, RASPA2-v2.0.47, LAMMPS-2021-9-29, and Zeo++-v0.3 are used in our experiments. Neural network modules are implemented with Py Torch-v1.11.0 (Paszke et al., 2019), Pyg-v2.0.4 (Fey & Lenssen, 2019), and Lightning-v1.3.8 (Falcon & The Py Torch Lightning team, 2019) with CUDA 11.3.
Experiment Setup Yes In our experiments, we use 3 rounds: U = 3, with σ = [3, 1.65, 0.3] and k = [30, 16, 1]. We use the Adam optimizer (Kingma & Ba, 2015) to maximize the model-predicted CO2 working capacity for 5,000 steps with a learning rate of 0.0003. All hyperparameters are reported in Table 3.