MHNN: Molecular Hypergraph Neural Networks

Conventional graphs only model the pairwise connectivity in molecules, failing to adequately represent higher-order connections like multi-center bonds and conjugated structures. To tackle this challenge, we introduce molecular hypergraphs and propose Molecular Hypergraph Neural Networks (MHNN) to predict molecular optoelectronic properties, where hyperedges represent conjugated structures.

🚀 Environment Setup

• We'll use conda to install dependencies and set up the environment. We recommend using the Python 3.9 Miniconda installer.
• After installing conda, install mamba to the base environment. mamba is a faster, drop-in replacement for conda:

  conda install mamba -n base -c conda-forge

• Create a new environment named mhnn and install dependencies.

  mamba env create -f env.yml

• Activate the conda environment with conda activate mhnn.

📌 Datasets

Dataset	Graphs	Task type	Task number	Metric
OPV	90,823	regression	8	MAE
OCELOTv1	25,251	regression	15	MAE
PCQM4Mv2	3,746,620	regression	1	MAE

OPV

The OPV dataset, named organic photovoltaic dataset, contains 90,823 unique molecules (monomers and soluble small molecules) and their SMILES strings, 3D geometries, and optoelectronic properties from DFT calculations. OPV has four molecular tasks, the energy of highest occupied molecular orbital for the monomer ($\varepsilon_{\rm HOMO}$), the lowest unoccupied molecular orbital of the monomer ($\varepsilon_{\rm LUMO}$), the first excitation energy of the monomer calculated with time-dependent DFT ($\Delta \varepsilon$), and the spectral overlap $I_{overlap}$. In addition, OPV has four polymeric tasks, the polymer $\varepsilon_{\rm HOMO}$, polymer $\varepsilon_{\rm LUMO}$, polymer gap $\Delta \varepsilon$, and optical LUMO $O_{\rm LUMO}$.

OCELOTv1

The OCELOTv1 dataset contains 25,251 organic $\pi$-conjugated molecules and their electronic, redox, and optical properties computed with the high accuracy DFT/TDDFT calculations. The DFT and TDDFT properties available in the dataset are vertical (VIE) and adiabatic (AIE) ionization energies, vertical (VEA) and adiabatic (AEA) electron affinities, cation (CR) and anion (AR) relaxation energies, HOMO energies (HOMO), LUMO energies (LUMO), HOMO–LUMO energy gaps (H–L), electron (ER) and hole (HR) reorganization energies, and lowest-lying singlet (S0S1) and triplet (S0T1) excitation energies.

PCQM4Mv2

PCQM4Mv2 is a quantum chemistry dataset originally curated under the PubChemQC project. A meaningful ML task was defined to predict DFT-calculated HOMO-LUMO energy gap of molecules given their 2D molecular graphs. PCQM4Mv2 is unprecedentedly large (> 3.8M graphs) in scale comparing to other labeled graph-level prediction datasets.

🔥 Model Training

OPV

1. We provide training scripts for MHNN and baselines under scripts/opv. For example, we can train MHNN for one task by running:

   bash scripts/opv/mhnn.sh [TASK_ID]

2. Train a model for all tasks by running:

   bash scripts/opv/run_all_tasks.sh [MODEL_NAME]

3. The OPV dataset will be downloaded automatically at the first time of training.

4. The model names and task ID for different tasks can be found here.

OCELOTv1

1. We provide training scripts for MHNN under scripts/ocelot. For example, we can train MHNN for one task by running:

   bash scripts/ocelot/train.sh [TASK_ID]

2. Train MHNN for all tasks by running:

   bash scripts/ocelot/run_all_tasks.sh

3. The ocelot dataset will be downloaded automatically at the first time of training.

4. Task ID for different tasks can be found here.

PCQM4Mv2

1. We provide a training script for MHNN under scripts/pcqm4mv2 to train MHNN by running:

   bash scripts/pcqm4mv2/train.sh

2. The PCQM4Mv2 dataset will be downloaded automatically at the first time of training.

🌈 Acknowledgements

This work was supported as part of NCCR Catalysis (grant number 180544), a National Centre of Competence in Research funded by the Swiss National Science Foundation.

📝 Citation

If you find our work useful, please consider citing it:

@article{chen2024molecular,
    author = {Chen, Junwu and Schwaller, Philippe},
    title = "{Molecular hypergraph neural networks}",
    journal = {The Journal of Chemical Physics},
    volume = {160},
    number = {14},
    pages = {144307},
    year = {2024},
    doi = {10.1063/5.0193557},
    url = {https://doi.org/10.1063/5.0193557},
}

📫 Contact

If you have any question, welcome to contact me at:

Junwu Chen: junwu.chen@epfl.ch