Raw VAE-Generated Polymer-Monomer Candidate Library

The dataset includes 2,000,000 generated polymer-monomer candidates represented as SMILES strings with polymerization-site markers ([*]). Among them, 1,800,738 were chemically valid, 1,995,121 were unique, and 1,994,891 were novel relative to the training set, corresponding to validity, uniqueness, and novelty values of 90.03%, 99.76%, and 99.99%, respectively. For each entry, the file records a unique sample identifier (sample_id), the generated structure (smiles), and three Boolean annotation fields indicating chemical validity (is_valid), uniqueness within the generated set (is_unique), and whether the structure is already present in the training set (in_training_set). This dataset constitutes the initial candidate pool for subsequent filtering, screening, and downstream validation.