基于深度学习的PD致病基因活性预测  

PREDICTION OF PD DISEASE GENE ACTIVITY BASED ON DEEP LEARNING

作　　者：李自臣[1,2] 田生伟 刘江越[1] 高双印[2] 

机构地区：[1]乌鲁木齐职业大学信息工程学院,新疆乌鲁木齐830002 [2]新疆大学软件学院,新疆乌鲁木齐830008

出　　处：《计算机应用与软件》2017年第9期183-187,共5页Computer Applications and Software

基　　金：新疆研究生科研创新基金项目(XJGRI2015034)

摘　　要：帕金森病PD(Parkinson’s disease)是一种神经性系统疾病,多发于中老年人。目前,该病情的病因和发病机制尚不明确,但根据多国临床试验数据统计与分析,PINKs基因是影响整个PD发病的重要原因之一。针对该基因的活性结构数据进行研究,提出基于深度学习的深度信念网络(DBN)与稀疏自编码(SAE)预测方法。该算法能通过深层网络特征单元自动学习到适合分类器分类的高层非线性组合特征,并将这些高层次特征输入到分类器中进行数据分析。实验结果表明,DBN算法的平均预测精度较SVM与ANN分别提高了28.04%、18.84%;SAE算法的平均预测精度较SVM与ANN分别提高了23.51%、14.31%。所以,提出的基于深度学习的PINKs活性预测方法具有较高的预测精度和稳定性,与理论分布也较为相吻合,适用于该基因活性的研究与探讨。Parkinson＇s disease（ PD） is a kind of nerve system disease,more common in the elderly. At present,the condition of the etiology and pathogenesis is not clear,but according to multinational clinical trial data statistics and analysis,PINKs gene is one of the important reason to influence the whole PD pathogenesis. This paper study for the structure of activity gene,and the DBN and SAE are proposed for the PINKs activity prediction. The proposed algorithm can learn automatically by the characteristics of deep web unit is suitable for the high nonlinear combination classifier classification feature,and will these high-level features inputs to the classifier for data analysis. The experimental results show that the DBN algorithm the average prediction accuracy of SVM with ANN respectively increased by 28. 04%,18. 84%; SAE algorithm the average prediction accuracy of the SVM and ANN respectively increased by 23. 51%,14. 31%. In this paper,based on the deep study of PINKs activity prediction method has higher prediction accuracy and stability,in conformity with the theory of distribution are,also is applicable to the activity of research and discussion.

关 键 词：活性 深度学习 SAE 预测 研究 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]