Dataset for: G-STAT: A Heterogeneous Graph Attention Framework for Explainable Genomic Variant Triage

This repository contains the processed heterogeneous graph dataset utilized in the manuscript "G-STAT: A Heterogeneous Graph Attention Framework for Explainable Genomic Variant Triage". The dataset represents a functionally dense 15:2 variant-gene tensor space, compiled from raw NCBI ClinVar VCF records utilizing SnpEff and dbNSFP (REVEL/CADD) annotations. The graph contains physical thermodynamic stress indicators, evolutionary constraints, and sparse topological gene hubs. Files Included: hetero_features.pt: The dense 15-dimensional variant and 2-dimensional gene feature matrices. hetero_edge_indices.pt: The bipartite edge topology connecting variants to their parent genes. hetero_labels.pt: Ground truth pathogenicity labels (Pathogenic, Benign, VUS). train_idx.pt / val_idx.pt: The strict zero-leakage training and validation masks used for architectural benchmarking. These PyTorch (.pt) files are formatted for immediate ingestion via PyTorch Geometric. Code to execute this data can be found at the associated GitHub repository.