Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Disentangling Hyperedges through the Lens of Category Theory
Authors: Yoonho Lee, Junseok Lee, Sangwoo Seo, Sungwon Kim, Yeongmin Kim, Chanyoung Park
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our proof-of-concept model experimentally showed the potential of the proposed criterion by successfully capturing functional relations of genes (nodes) in genetic pathways (hyperedges). Our implementation is available at https://github.com/Yoonho-Lee-AI4Science/Natural-HNN. |
| Researcher Affiliation | Academia | Yoonho Lee KAIST EMAIL Junseok Lee KAIST EMAIL Sangwoo Seo KAIST EMAIL Sungwon Kim KAIST EMAIL Yeongmin Kim KAIST EMAIL Chanyoung Park KAIST EMAIL |
| Pseudocode | No | The paper describes the steps for its model, Natural-HNN, in Section 4, but it does not present them in a formally labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | Our implementation is available at https://github.com/Yoonho-Lee-AI4Science/Natural-HNN. |
| Open Datasets | Yes | For the cancer subtype classification task, we downloaded clinical data for 6 cancer types (BRCA, STAD, SARC, LGG, CESC, HNSC) and preprocessed data following Pathformer [46] (Details in Appendix B.2). We downloaded pathways from several pathway databases including KEGG [33], PID [61], Reactome [8] and Biocarta.[55]. For BRCA and STAD, we gathered cancer subtypes from TCGA [73] using TCGAbiolinks [7, 64, 53] R library. We gathered m RNA/mi RNA expression, DNA methylation, DNA copy number variation (CNV) using TCGAbiolinks. The results shown here are in whole or part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. |
| Dataset Splits | Yes | We randomly split the data into 50%/25%/25% for training/validation/test set. We measured average and standard deviation of the performances for 10 different data splits. For the node classification task with standard hypergraph benchmark datasets, we randomly split the data into 50%/25%/25% for training/validation/test set. |
| Hardware Specification | Yes | We used 48GB NVIDIA RTX A6000 GPU. |
| Software Dependencies | Yes | We created a anaconda environment with python 3.7.16, pytorch 1.11.0 and pytorch geometric with version 2.0.4. |
| Experiment Setup | Yes | The hyperparameter search space is provided in Appendix C.5. We used Adam optimizer for Natural-HNN. We fixed the number of layers to 2, except for HSDN, because HSDN uses only a single layer. During training, we set 50 as the batch size. Generally, we used 0.5 as dropout. |