Unspecified QM9NMR train/test split used by DetaNet

Ascertain the specific training and test split of the QM9NMR dataset used by DetaNet, including the exact molecule assignments to each split, to ensure reproducible benchmarking and fair comparison of predictive performance.

Background

For benchmarking on QM9NMR, the authors follow recent settings but indicate a reproducibility gap: they were unable to locate the exact train/test partition used in DetaNet’s reported results. They reconstructed a split from supplementary information, which may not perfectly match the original division.

Clarifying the official split is important to enable direct, apples-to-apples comparisons between models and to ensure that improvements reported by NMRNet or future methods are attributable to modeling advances rather than differences in dataset partitioning.