The dataset for Graph-Based Construction of the Astrochemical Molecular Inventory toward Sgr B2(N)

This Zenodo repository provides the dataset associated with the study “Graph-Based Construction of the Astrochemical Molecular Inventory toward Sgr B2(N)”. The data are divided into two folders, testing_data and total_data, each containing 20 files named A01 through A20, representing the 20 subregions of Sgr B2(N) analyzed in this work. testing_data contains the held-out testing records for each subregion. For each sample, the file provides the subregion label, sample ID, raw testing input features, original testing label, ground-truth column density, and the corresponding predictions from four regression models: Random Forest, Linear Regression, Gradient Boosting Regression, and Gaussian Process Regression. The testing features are constructed from the physical and chemical descriptors used in the manuscript, including source size, rotational temperature, line width, velocity offset, vibrational state, and SELFIES-based molecular representation. total_data contains the full set of molecule-condition records for each subregion. Each entry includes the sample ID, source size, rotational temperature, line width, velocity offset, vibrational state, subregion tag, SELFIES representation, column density, molecule name, and canonical SMILES string. This folder therefore provides the complete molecular inventory used for the subregion-wise modeling and interpretation analyses. All files are distributed in tab-separated text format. The dataset is provided to facilitate transparency, reproducibility, and direct comparison between observed and predicted molecular column densities across the 20 subregions.