What problem does it solve? Running interpretable graph neural networks on brain connectome data requires careful handling of PyG graph formats, edge-weight normalization, and task configuration, which this Skill packages into a reproducible training workflow for fMRI phenotype prediction. ## Core Features & Use Cases - IBGNN Training Pipeline: Trains the MPConv-based IBGNN model with learnable MLP message functions over [x_i, x_j, edge_attr] using K-fold cross-validation. - Classification and Regression Support: Handles both classification (e.g., gender prediction) and regression (e.g., HCP age prediction) through a unified interface with nclass and task flags. - BrainGNN Data Reuse: Loads preprocessed PyG Data graphs from data/braingnn_input/<atlas>/ via the shared NeuroClawFCDataset, avoiding duplicate preprocessing. - Use Case: A researcher with HCP functional connectivity matrices wants to predict subject age using an interpretable GNN; they run the reference training script with the schaefer_100_7net atlas and a labels CSV to get MAE metrics across folds. ## Quick Start Train IBGNN on the aal_116 atlas with my HCP gender labels CSV for 10 epochs on fold 0 as a smoke test.
