DHG-LGB v2.0.0 - Complete Data, Code, and Results

DHG-LGB v2.0.0 - Complete Archive Complete implementation of Disease-Hypergraph integrated with LightGBM framework for metabolite-disease association prediction. This archive provides all data, code, and experimental results to ensure full scientific reproducibility, addressing reviewer requirements for permanent data availability. CONTENTS: Data (442 MB)- Raw data: HMDB 5.0, CTD (2023), protein sequences, Gene Ontology, SMILES structures- Processed data: Pre-computed similarity matrices, hypergraph structure (178 diseases × 19,442 nodes), pre-trained HGNN embeddings (saves 2 days GPU time) Source Code (111 KB)- HGNN implementation, LightGBM classifier, preprocessing and evaluation scripts- Complete documentation and installation instructions Experimental Results (53 KB)- Table 2: Performance comparison of 6 classifiers- Table 3: 95% confidence intervals for MCC, AUC, AUPRC- Table 4: Robustness under imbalanced ratios (1:1 to 1:10)- Detailed 5-fold cross-validation results- Case studies: Obesity, Schizophrenia, Crohn's Disease- Baseline comparisons: PageRank, KATZ, EKRR, GCNAT, MDA-AENMF FEATURES:- Complete reproducibility at three levels- Pre-computed embeddings enable reproduction in 10 minutes vs 2+ days- 8 detailed README files- FAIR compliant TECHNICAL DETAILS:- Entities: 2,006 metabolites, 4,912 proteins, 12,524 GO terms, 178 diseases- Associations: 4,000 validated metabolite-disease pairs- Performance: AUC=0.9983±0.0001, MCC=0.9305±0.0012 CITATION:Xiao F, Ran Y, Li Z. Identifying Metabolite-Disease Associations via Messaging in Hypergraphs. Metabolites. 2025. LICENSE: MIT GitHub Repository: https://github.com/xavierxiao848/DHG-LGB-Repository For questions, open an issue on GitHub or contact the corresponding author.