计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (10): 98-103.

• 研发、设计、测试 • 上一篇    下一篇

基于概率神经网络的蛋白质相互作用分类器

张 伟1,2,郝江锋3,项俊平4,5,胡茂林4,5   

  1. 1.中国矿业大学 徐海学院 数理教研室,江苏 徐州 221008
    2.中国矿业大学 理学院,江苏 徐州 221008
    3.巢湖学院 数学系,安徽 巢湖 238000
    4.安徽大学 计算智能与信号处理教育部重点实验室,合肥 230039
    5.安徽大学 数学与计算科学学院,合肥 230039
  • 收稿日期:2007-08-13 修回日期:2007-12-06 出版日期:2008-04-01 发布日期:2008-04-01
  • 通讯作者: 张 伟

Predicting type of protein-protein interaction based on probabilistic neural network

ZHANG Wei1,2,HAO Jiang-feng3,Xiang Jun-ping4,5,HU Mao-lin4,5   

  1. 1.Faculty of Mathematics,Xuhai College,China University of Mining Technology,Xuzhou,Jiangsu 221008,China
    2.College of Science,China University of Mining Technology,Xuzhou,Anhui 221008,China
    3.Department of Mathematics,Chaohu College,Chaohu,Anhui 238000,China
    4.Key Laboratory of Intelligent Computing & Signal Processing,Ministry of Education,Anhui University,Hefei 230039,China
    5.School of Mathematics & Computational Science,Anhui University,Hefei 230039,China
  • Received:2007-08-13 Revised:2007-12-06 Online:2008-04-01 Published:2008-04-01
  • Contact: ZHANG Wei

摘要: 蛋白质-蛋白质作用面上的结构特征对于研究蛋白质功能具有重要意义。提出了一种新的、基于统计直方图提取蛋白质作用面特征的方法,并且利用提取出的作用面特征,结合概率神经网络,实现了对作用面结构类型的分类预测。从预测结果来看,统计直方图提取出的特征,对蛋白质作用面结构具有很好的区分能力,而且可以通过调节划分的区间个数和节点的选取方式,达到对作用面结构的不同粒度的描述,以适用于不同目的的研究,这可能对与结构有关的某些生物信息学问题的研究具有启发性。利用概率神经网络对作用面结构进行分类预测,避开了费时的结构比对和数据库搜索,且训练快速,扩展能力强,正确率高,对独立测试集的911个蛋白复合物视在正确率达到90.67%。基于该算法的MATLAB分类器软件可以通过E-Mail与作者联系获取。

关键词: 统计直方图, 蛋白质作用面, 概率神经网络, 结构比对, 分类器

Abstract: In order to predict the structure type of Protein-Protein Interface(PPI),a new interface structure feature generation method using histogram is proposed.The interface can be described on diverse level of granularity by adjusting the number of parts and the partition nodes of histogram.From the result of prediction,which is based on Probabilistic Neural Network(PNN), the histogram method has a good ability to distinguish the different structure type of PPI,which may have a potential value for some bioinformatics problem related to protein structure.By using PNN,a training-fast neural network,the prediction avoids the structure alignment and database searching,which is very time-consuming.The Histogram-PNN classification machine has correct rate of 90.67% on testing set containing 911 protein complexes.

Key words: histogram, Protein-Protein Interface(PPI), Probabilistic Neural Network(PNN), structure alignment, classifier