Substructure Aware Graph Neural Networks

Authors: DingYi Zeng, Wanlong Liu, Wenyu Chen, Li Zhou, Malu Zhang, Hong Qu

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our extensive experiments demonstrate the effectiveness of our framework, achieving state-of-the-art performance on a variety of well-proven graph tasks, and GNNs equipped with our framework perform flawlessly even in 3-WL failed graphs. Specifically, our framework achieves a maximum performance improvement of 83% compared to the base models and 32% compared to the previous state-of-the-art methods. Our extensive experiments validate the state-of-the-art performance of our framework on various base model networks, tasks and datasets, especially on the graph regression task of drug constrained solubility prediction (ZINCFULL). Our framework achieves a maximum MAE reduction of 83% compared to the base model and a maximum MAE reduction of 32% compared to the previous state-of-the-art model.
Researcher Affiliation Academia 1School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China
Pseudocode No No pseudocode or algorithm blocks were found explicitly labeled or formatted as such in the paper.
Open Source Code Yes Our implementation is available at https://github.com/ Black Halo-Drake/SAGNN-Substructure-Aware-Graph-Neural Networks
Open Datasets Yes We use EXP (Abboud et al. 2021) and SR25 (Balcilar et al. 2021) datasets to verify the expressiveness improvement of our framework for MPNNs, where EXP contains 600 pairs of 1-WL failed graphs and SR25 contains 15 3-WL failed strongly regular graphs. For performance on real-world tasks, we use three kinds of datasets of different scales for validation. The first kind is small-scale real-world datasets TUDataset (Morris et al. 2020), which includes MUTAG, PTC, PROTEINS, NCI1 and IMDB from biology, chemistry and social networks. The second is a large-scale molecular benchmark from the zinc database from the ZINC database (Sterling and Irwin 2015), which includes ZINC (12K graphs) (Dwivedi et al. 2020) and ZINC-FULL (250k graphs) (G omez-Bombarelli et al. 2018; Jin, Barzilay, and Jaakkola 2018; Zhang et al. 2018a). The last is a molecular large-scale dataset from the Open Graph Benchmark (Hu et al. 2020, 2021) which includes OGBG-MOLHIV (41k graphs) and OGBG-PCBA (437k graphs).
Dataset Splits Yes For TUDataset, we perform 10-fold cross-validation and report the average and standard deviation of validation accuracy across the 10 folds within the cross-validation.
Hardware Specification No The paper does not explicitly mention specific hardware components (e.g., GPU models, CPU types, memory) used for running the experiments.
Software Dependencies No The paper does not explicitly list specific software dependencies with their version numbers (e.g., Python version, library versions) needed for replication.
Experiment Setup Yes Hyperparameter effect We first conduct an ablation study to analyze the impact of the values of Ego and Cut on model performance on different tasks and different data schema. Figure 4 visualizes the results of our hyperparameter experiments on three datasets. All three datasets show the same performance trend, that is, the best performance in the case of Ego=3, Cut=4, which shows our model s great hyperparameter stability. Experiments are repeated 10 times on expressiveness datasets and 3 times on large scale datasets to calculate mean and standard derivation. For TUDataset, we perform 10-fold cross-validation and report the average and standard deviation of validation accuracy across the 10 folds within the cross-validation.