Comment on “Hybrid Graph–Machine Learning Framework for Accurate and Interpretable Band Gap Prediction”
Aydin et al. reported R2 = 0.921 for band gap prediction using hybrid graph neural networks. Reproduction shows this performance is inflated by data leakage from post-DFT proxies (is_metal, is_gap_direct) and embedding contamination. Controlled ablations reveal a genuine GNN gain of 0.08–0.11 in R2, while leakage contributes ∼0.06, lowering honest performance to 0.82–0.86 R2 (MAE 0.28–0.30 eV).
Authors
- Shambhu Bhandari Sharma (ORCID: https://orcid.org/0000-0002-6003-7425)
Institutions
- University College London (GB)
Publication Details
- Journal
- Journal of Chemical Information and Modeling
- Published
- 2026-11-19
- DOI
- https://doi.org/10.1021/acs.jcim.6c03192
- Primary Topic
- Machine Learning in Materials Science
- Type
- article
- Field-Weighted Citation Impact
- 0.00