QMOF-derived sparsified atomic graph views generated by ChemSPAR

This dataset contains QMOF-derived sparsified atomic graph views generated for the manuscript “Atomic Black Hole Sparsification for Chemistry-Aware Frugal Learning on QMOF Graphs.”

The dataset provides derived graph-view representations of QMOF structures for chemistry-aware graph sparsification experiments. Each MOF retains its original crystal identity, composition, and target label, while the computational atom–atom edge set is modified according to Atomic Black Hole (ABH), CrystalNN-calibrated CD-ABH, and baseline sparsification methods.

The release includes original over-complete atomic graph views constructed using a 4.5 Å distance cutoff, ABH-sparsified graph views at multiple pruning thresholds, CD-ABH graph views, baseline pruning outputs, fixed train/validation/test split information, edge-level audit logs, benchmark tables, configuration files, and documentation for reproducibility.

This is not a new raw MOF structure dataset. Raw QMOF structures are not redistributed in this package and should be obtained from the original QMOF source under its terms of use. This release provides reproducible, QMOF-derived graph-view datasets and audit/benchmark files to evaluate chemistry-aware, auditable graph sparsification in materials informatics.

Associated code repository:
https://github.com/MehrdadJalali-AI/Atomic_Black_Hole