AcetylBench

AcetylBench is a deduplicated benchmark dataset for lysine acetylation site prediction, curated from the dbPTM database using ProteinBERT embedding-based cluster purification followed by CD-HIT sequence deduplication at 40% pairwise identity threshold. Dataset contents (this record): AcetylBench_train_positive.csv — 27,182 acetylated lysine sequence windows (31-mer) AcetylBench_train_negative.csv — 7,063 non-acetylated lysine sequence windows (31-mer) Pre-computed ESM-1v embeddings (mean pool and lysine-position representations) for three dataset splits: AcetylBench training set, DeepACET training set, and independent test set (12 .npy files) Analysis code: https://github.com/uzairamu/AcetylBench