排序学习前向掩蔽模型在T细胞表位预测中的应用  

Applications of sequential learning ahead masking model to T cell epitope prediction

作　　者：曾安[1] 潘丹[2] 郑启伦[3] 彭宏[3] 

机构地区：[1]广东工业大学计算机学院,广东广州510006 [2]中国移动通信集团广东有限公司,广东广州510100 [3]华南理工大学计算机科学与工程学院,广东广州510640

出　　处：《计算机应用》2007年第1期80-83,共4页journal of Computer Applications

基　　金：广东省自然科学基金资助项目(6300252);广东工业大学博士启动基金资助项目(063001);国家自然科学基金资助项目(30230350)

摘　　要：在综述了T细胞表位预测的定义,意义和研究现状的基础上,分析了当前流行的基于误差反向传播前馈神经网络(BPNN)的T细胞表位预测模型的不足,即网络结构较难确定、训练速度慢和难以增量学习等,提出了利用排序学习前向掩蔽(SLAM)模型及其增量学习算法作为T细胞表位预测方法,并给出了构建T细胞表位预测模型的基本步骤。基因HLA-DR4(B1*0401)编码的MHC II类分子结合肽的应用实例表明,与基于BPNN的T细胞表位预测模型相比,基于SLAM的T细胞表位预测模型不但能在极短时间内完成样本的学习,而且能有效地实现增量学习。The definition, the meaning and the state-of-art of T cell epitope prediction were firstly summarized. And then, the disadvantages of the prevailing T cell epitope prediction model based on the Back-Propagation Neural Networks （BPNN）, including difficulties in presetting networks structure, converging and incremental learning, were investigated. In terms of the above-mentioned drawbacks, Sequential Learning Ahead Masking model （SLAM） and its fast incremental learning algorithm were deliberately chosen to predict T cell epitope. Meanwhile, the basic steps of constructing T cell epitope prediction model based on SLAM were advocated. Finally, a case study of predicting the binding capacities to MHC class II molecule encoded by gene HLA-DR4 （ B1 ＊ 0401.） was given in detail. The application results show that T cell epitope prediction model based on SLAM has better learning performance and stronger incremental learning capabilities than that based on conventional BPNN.

关 键 词：T细胞表位预测 排序学习 神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP183[自动化与计算机技术—控制科学与工程]