为了提高抗癌药物的发现效率并降低研发成本，针对基于PLK1此类结构和功能均高度保守的丝氨酸／苏氨酸蛋白激酶在多种肿瘤类型中高表达的特点，提出以PLK1 PBD为靶点的深度信念网络（deep believe network，DBN）抗癌活性研究方法。利用深度学习思想，对20000个化合物的抗癌活性进行分析，并分别与ANN、SVM方法进行对比验证。实验结果表明，在同等条件下，DBN网络针对抗癌药物活性研究具有突出的优势，其平均预测活性的精确度可达91．05％，明显高于ANN和SVM，从而实现了对化合物抗癌活性的良好评估。%For reducing research costs and improving the efficiency of drug discovery,based on the class of highly conserved serine/threonine protein kinases,which is overexpressed in a wide spectrum of cancer types.In order to PLK1 PBD as target, this paper proposed a prediction method of using deep belief network and took advantage of the related ideas of deep learning to analysis 20 000 compounds of antitumor activity,compared with ANN and SVMfor verifying.The experimental results show that DBN network has outstanding advantages at the same conditions,the average prediction accuracy can get to 9 1 .05%,which is higher than the method of ANN and SVM obviously,applies to evaluate anticancer activity.
展开▼
