We introduce LG-Mol, a novel framework that seamlessly integrates local and global information. It employs a tailored Transformer architecture for atomic feature extraction and a graph neural network with equivariance properties to analyze the electrostatic potential surface.

LG-Mol is built upon the foundation of Uni-Mol. Therefore, you need to install Uni-Mol first. For instructions on installing Uni-Mol, please refer to the Uni-Mol installation guide. After installing Uni-Mol, move the files from the unimol folder (in this repository) to the corresponding folder within the installed Uni-Mol directory.

We provide the code for data preprocessing in the data folder. The preprocessed data is also available here.

The scripts required to run LG-Mol on the MoleculeNet benchmark are provided in the scripts folder. Here is an example:

bash molecular_property_prediction_bbbp_h.sh (fill up you own data path first)