ATTN-Lab/bbbnuke-screening-1B

BBB-Nuke数据集是一个用于血脑屏障渗透性筛选的大规模化学数据集，包含10.2亿个小分子化合物。该数据集通过BBB-Nuke管道（v0.12.0）进行筛选，涵盖标准化（使用RDKit进行SMILES规范化）、理化性质计算（如分子量、LogP、TPSA等）、pKa预测（使用MolGpKa）、CNS-MPO评分（多参数优化）、外排转运蛋白预测（针对8种蛋白质）以及分类器和启发式方法（最终生成BBB渗透概率评分，5折交叉验证AUROC为0.933）。数据来源包括Enamine REAL（约1.99亿化合物）、现有数据（如ZINC/ChEMBL，约6.66亿）、PubChem（约5900万）和CPU筛选数据（约9900万）。关键特征包括标准化SMILES、BBB渗透概率（p_bbb）、CNS-MPO评分、理化描述符、预测pKa值、外排转运蛋白亲和力评分以及数据来源标识。该数据集适用于药物发现、化学信息学和血脑屏障相关研究，遵循CC BY-NC 4.0许可证。BBB-Nuke: 1 Billion Compound Blood-Brain Barrier Permeability Screen is a dataset of 1.02 billion small molecules screened for blood-brain barrier (BBB) permeability using the BBB-Nuke pipeline (v0.12.0). The screening process includes standardization (SMILES canonicalization via RDKit), physicochemical property calculation (e.g., MW, LogP, TPSA), pKa prediction (via MolGpKa), CNS-MPO scoring, efflux transporter prediction (for 8 proteins), and a classifier with heuristics to generate a final BBB permeability probability score (with 5-fold CV AUROC 0.933). Data sources include Enamine REAL (~199M compounds), existing data (e.g., ZINC/ChEMBL, ~666M), PubChem (~59M), and CPU screen data (~99M). Key columns cover standardized SMILES, BBB permeability probability (p_bbb), CNS-MPO score, physicochemical descriptors, predicted pKa, efflux transporter affinity scores, and data provenance. The dataset is intended for drug discovery, cheminformatics, and BBB-related research, licensed under CC BY-NC 4.0.