KA-GNN

This is the code of KA-GNN

Table of Contents

1. Environment Requirements
2. Installation Steps
3. Data Download and Configuration
4. Running the Project
5. Experimental Results

Environment Requirements

This project requires:

• Python Version: 3.11
  • Use Python 3.11, as this is the version used for development and testing of the code.
• CUDA Version: 11.7
  • To fully utilize GPU acceleration, ensure that your environment supports CUDA 11.7.

Installation Steps

1. Create a Virtual Environment

Recommended to use conda:

conda create -n myenv python=3.11
conda activate myenv

2. Install Dependencies

Install the necessary Python libraries from requirements.txt:

pip install -r requirements.txt

3. Verify CUDA Installation

Check that CUDA 11.7 is correctly installed on your system:

nvcc --version

Data Download and Configuration

Download Data

Download the datasets from MoleculeNet. Place the datasets into the data directory in your project folder data.

Configure Dataset Usage

To use different datasets, modify the c_path.yaml file in the config directory:

select_dataset: "bace"  # Replace "bace" with "hiv" or "muv" as needed

Running the Project

Execute the project with the configured dataset by running:

python main.py

KA-GNNs(KA-GCN and KA-GAT)

Experimental Results

Comparison with Non-Pretrained Models

The following table presents the comparison of KA-GNN with various GNN architectures. The best performance values are highlighted in bold, and standard deviation values are indicated in subscripts.

Model	BACE	BBBP	ClinTox	SIDER	Tox21	HIV	MUV
No. mol	1513	2039	1478	1427	7831	41127	93808
No. avg atoms	65	46	50.58	65	36	46	43
No. tasks	1	1	2	27	12	1	17
D-MPNN	0.809_(0.006)	0.710_(0.003)	0.906_(0.007)	0.570_(0.007)	0.759_(0.007)	0.771_(0.005)	0.786_(0.014)
AttentiveFP	0.784_(0.022)	0.663_(0.018)	0.847_(0.003)	0.606_(0.032)	0.781_(0.005)	0.757_(0.014)	0.786_(0.015)
N-GramRF	0.779_(0.015)	0.697_(0.006)	0.775_(0.040)	0.668_(0.007)	0.743_(0.009)	0.772_(0.004)	0.769_(0.002)
N-GramXGB	0.791_(0.013)	0.691_(0.008)	0.875_(0.027)	0.655_(0.007)	0.758_(0.009)	0.787_(0.004)	0.748_(0.002)
PretrainGNN	0.845_(0.007)	0.687_(0.013)	0.726_(0.015)	0.627_(0.008)	0.781_(0.006)	0.799_(0.007)	0.813_(0.021)
GROVE_base	0.821_(0.007)	0.700_(0.001)	0.812_(0.030)	0.648_(0.006)	0.743_(0.001)	0.625_(0.009)	0.673_(0.018)
GROVE_large	0.810_(0.014)	0.695_(0.001)	0.762_(0.037)	0.654_(0.001)	0.735_(0.001)	0.682_(0.011)	0.673_(0.018)
GraphMVP	0.812_(0.009)	0.724_(0.016)	0.791_(0.028)	0.639_(0.012)	0.759_(0.005)	0.770_(0.012)	0.777_(0.006)
MolCLR	0.824_(0.009)	0.722_(0.021)	0.912_(0.035)	0.589_(0.014)	0.750_(0.002)	0.781_(0.005)	0.796_(0.019)
GEM	0.856_(0.011)	0.724_(0.004)	0.901_(0.013)	0.672_(0.004)	0.781_(0.001)	0.806_(0.009)	0.817_(0.005)
Mol-GDL	0.863_(0.019)	0.728_(0.019)	0.966_(0.002)	0.831_(0.002)	0.794_(0.005)	0.808_(0.007)	0.675_(0.014)
Uni-mol	0.857_(0.002)	0.729_(0.006)	0.919_(0.018)	0.659_(0.013)	0.796_(0.005)	0.808_(0.003)	0.821_(0.013)
SMPT	0.873_(0.015)	0.734_(0.003)	0.927_(0.002)	0.676_(0.050)	0.797_(0.001)	0.812_(0.001)	0.822_(0.008)
KA-GCN	0.890_(0.014)	0.787_(0.014)	0.989_(0.003)	0.842_(0.001)	0.799_(0.005)	0.821_(0.005)	0.834_(0.009)
KA-GAT	0.884_(0.004)	0.785_(0.021)	0.991_(0.005)	0.847_(0.002)	0.800_(0.006)	0.823_(0.002)	0.834_(0.010)

Citing

If you find our codes useful in your research, please consider citing:

@article{li2025kolmogorov,
  title={Kolmogorov--Arnold graph neural networks for molecular property prediction},
  author={Li, Longlong and Zhang, Yipeng and Wang, Guanghui and Xia, Kelin},
  journal={Nature Machine Intelligence},
  pages={1--9},
  year={2025},
  publisher={Nature Publishing Group}
}