Data for the manuscript entitled "Interpreting graph neural networks reveals microbial association patterns in inflammatory bowel diseases"

Processed data and results accompanying the manuscript by Cunha et al. The code is at github.com/Multiomics-Analytics-Group/IBD_Individual_MANs. How to use Unzip every zip file into the root of the code repository. Paths inside the zip files start with data/ or results/, which are the locations the scripts expect (see code/common/paths.R and code/common/paths.py). bash:for z in *.zip; do unzip -o "$z" -d /path/to/IBD_Individual_MANs; done Contents Zip file Unpacks to What it holds data_raw.zip data/raw/curatedMetagenomicData/ Species-level estimated read counts and sample metadata as downloaded from curatedMetagenomicData (.rds, R lists keyed by study) data_abundance.zip data/processed/abundance/ Prevalence-filtered count tables, with and without metadata (.rds), and the HMP2 and Hall tables as .csv data_networks.zip data/processed/networks/ Per-individual SparCC association matrices (.rds) and the unthresholded individual-level microbial association networks (.gml): 106 for HMP2, 25 for Hall data_pyg_datasets.zip data/processed/pyg_datasets/ PyTorch Geometric datasets at association thresholds 0.3, 0.4 and 0.5 (.pt), the matching CLR mean, CLR variance and taxonomy tables (.csv), and the thresholded networks (.gml) results_models.zip results/models/ Per-run metrics, ROC and precision-recall curves, predictions and selected features for the graph-based and traditional models (.csv) results_interpretation.zip results/interpretation/ Integrated Gradients node and edge attributions for all 50 model instances.(.csv) results_ablation.zip results/ablation/gnn/ Predictions, metrics and trained models of the metagraph ablation experiments, and SparCC weights among metagraph Bacteroidales (.csv, .pt) Individuals are identified by their curatedMetagenomicData subject IDs. In the model files, non-IBD is encoded as the positive class. Notes - Loading the .pt files: they store a custom dataset class, so add code/common from the code repository to the Python path before calling torch.load(path, weights_only=False). References - Lloyd-Price J. et al. Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases. Nature 569, 655–662 (2019).- Hall A. B. et al. A novel Ruminococcus gnavus clade enriched in inflammatory bowel disease patients. Genome Medicine 9, 103 (2017).- Pasolli E. et al. Accessible, curated metagenomic data through ExperimentHub. Nature Methods 14, 1023–1024 (2017).