基于深度信念网络的CYP450 2C9抑制性分类

细胞色素P450 2C9 (Cytochrome P450 2C9)是人体肝脏中重要的代谢酶,参与多种药物代谢,约占CYP450蛋白总量的15%～20%。利用深度学习思想,提出基于深度信念网络DBN (Deep Belief Network)的CYP450 2C9抑制性分类模型。实验选用13 000个化合物作为数据集,采用Pub Chem和MACCS分子指纹进行分子结构表征。利用DBN的半监督学习方式从预处理后的特征中学习更本质的特征表示,避免人工提取特征的过程,实现CYP450 2C9的抑制性分类。实验结果表明:在同等条件下,DBN相比于SVM和ANN具有明显优势,平均分类准确率为80.6%,灵敏度(SE)为86.9%,特异性(SP)为66.2%,对药物筛选和新药研发具有积极意义。

CLASSIFICATION OF CYP450 2C9 INHIBITORS BASED ON DEEP BELIEF NETWORK

The cytochrome P450 2C9 is one of the important metabolic enzymes in the human liver, which involves in many kinds of drug metabolism and accounts for about 15%~20% of the total CYP450 protein. Deep learning was adopted to propose a CYP450 2C9 inhibitory classification model based on deep belief network (DBN). In the experiment, 13 000 compounds were selected as data sets. PubChem and MACCS molecular fingerprints were used to characterize the molecular structure. We used DBN semi-supervised learning method to learn more essential feature representation from the features obtained by pre-processing, thus avoiding the process of extracting features manually and realizing classification of CYP450 2C9 inhibitors. Experimental results show that DBN has outstanding advantages compared with SVM and ANN at the same conditions. The average accuracy is 80.6%, the sensitivity (SE) is 86.9% and the specificity (SP) is 66.2%. It has a positive significance in drug screening and development.