基于机器学习的抗胰岛β细胞凋亡化合物分类模型研究

Machine Learning-Based Classification Model of Anti-Pancreatic β-Cell Apoptosis Compounds

摘要: 糖尿病是一种以胰岛素抵抗和胰岛细胞功能衰竭为特征的慢性代谢性疾病，现有的胰岛素注射和口服降糖药治疗方法均无法从根本上逆转糖尿病的原发性病理机制，故针对胰岛β细胞受损干预，寻找抑制β细胞凋亡的药物将对糖尿病的治疗具有重大意义。基于已报道的103个具有抗胰岛β细胞凋亡活性化合物，结合6种分子指纹和7种机器学习方法对上述化合物构建了分类模型，并运用10倍交叉验证和测试集对模型性能进行评估。结果表明，基于PubChem指纹的随机森林算法所构建的模型表现最佳(AUC = 0.992、CA = 0.96和MCC = 0.901)。同时结合信息增益和子结构频率分析，识别出具有抗胰岛β细胞凋亡活性的9个特征子结构，如含氮芳香杂环、胺类和吡啶等高活性片段，为抗胰岛β细胞凋亡化合物的开发提供理论参考和指导。

Abstract: Diabetes is a chronic metabolic disease characterized by insulin resistance and pancreatic β-cell dysfunction. Current treatments, including insulin injections and oral hypoglycemic drugs, cannot fundamentally reverse the primary pathological mechanisms of diabetes. Therefore, targeting interventions that protect pancreatic β-cells and identifying drugs that inhibit β-cell apoptosis are of significant importance for diabetes treatment. Based on 103 reported compounds with anti-β-cell apoptosis activity, a classification model was constructed using six molecular fingerprints and seven machine learning methods. The model’s performance was evaluated using 10-fold cross-validation and a test set. The results show that the model built using the Random Forest algorithm based on PubChem fingerprints performed the best (AUC = 0.992, CA = 0.96, and MCC = 0.901). Additionally, through information gain and substructure frequency analysis, nine characteristic substructures with anti-β-cell apoptosis activity were identified, including nitrogen-containing aromatic heterocycles, amines, and pyridines. These findings provide theoretical reference and guidance for the development of anti-β-cell apoptosis compounds.