Fragment-Grounded Neural Simulation of Electron Ionization Mass Spectra at Library Scale – Supplementary Data

Abstract

Predicted electron-ionization mass spectra (GC-EI-MS) for 87,260,988 molecules from a filtered PubChem library, generated with ICICLE (fragment-grounded neural EI-MS simulator; model trained on the NIST 2023 GC-MS random split). Each entry includes canonical SMILES, InChIKey-14, nominal molecular weight, and a predicted spectrum binned at 1 Da resolution (0–750 m/z). Intended for large-scale spectral library search / retrieval benchmarking in small-molecule mass spectrometry. Data is split into 10 row-range HDF5 parts (part00–part09); concatenate in order to reconstruct the full dataset. See accompanying README for the exact PubChem filtering funnel and provenance.