PhyChiral Dataset and Reproducibility Artifacts for Automated Chiral Crystal Recognition

This dataset supports the ACS Nano manuscript “A Physics-Constrained Explainable Artificial Intelligence Framework for Automated Chiral Crystal Recognition”. DATA CONTENTS• 86 original Asp-SEM images.• A frozen 72-image Asp-SEM development source pool (32 D and 40 L).• Nested-cross-validation split definitions and frozen evaluation artifacts.• 588 historical augmented training images (297 D and 291 L).• Cross-modality and cross-compound collections: 34 Asp-OM images, 91 Tyr-SEM images, and 73 Tyr-OM images.• SHA-256 file manifest and evidence-boundary documentation. ANALYSIS SCOPEDINOv2 C-M-A-Q is the primary analysis pipeline. EfficientNet is retained only as a supplementary baseline. The reported 90.28% accuracy is a same-cohort post hoc development estimate obtained from the same 72-image Asp-SEM source cohort; it is not an independent prospective validation result. DATA DISTRIBUTION AND INTEGRITYVersion 2.0.0 contains four complete ZIP archives and two byte-split logical ZIP sets for the augmented D and L collections. Download every .partNNN file together with zenodo_parts_manifest.json and reassemble_augmented_archives.py, then run python reassemble_augmented_archives.py . to reconstruct PhyChiral_Data_v2_aug_D.zip and PhyChiral_Data_v2_aug_L.zip. Extract all six logical ZIP archives into the same destination directory; they share the PhyChiral_Data_v2/ root and contain disjoint payload files. The complete extracted-file SHA-256 inventory is MANIFEST_SHA256.csv in PhyChiral_Data_v2_core.zip. zenodo_parts_manifest.json SHA-256: 77e6ca377d851fe197feab7f351b4984b1bbbb7c99eb0d3b9b9eefd5b1242817. Reassembly helper SHA-256: ef50202f4e5fa8ca6801c9b2228caf2748ad3d03e957c9ba00d05d7c225daa80. CODEThe corresponding implementation, frozen predictions, fold definitions, tests, and reproducibility instructions are available at https://github.com/queendio/Automated-Chiral-Recognition.