【作者】 刘涛； 宋哲； 刘伟；

【Author】 LIU Tao, SONG Zhe, LIU Wei (Department of Physics, Dalian university of technology, Dalian 116024,China)

【机构】 大连理工大学物理系；

【摘要】 由于实验上细胞毒性T细胞(cytotoxicity T lymphocytek,CTL)表位预测方法操作繁琐且费时费力耗资大,因此各种CTL表位的理论预测方法应运而生,人工神经网络方法就是主要的一种。本文介绍了基于神经网络的CTL表位预测模型,作出附加动量项和自适应学习率等两点改进。结果表明:改进后的模型具有训练快,预测准的特点。文章利用大量的表位数据进行了训练预测,给出了本模型的性能数据并计算得到的一套最优的参数。通过与已有结果的比较,表明该模型对预测CTL表位能给出更为满意的结果。更多还原

【Abstract】 Because the experimental methods of predicting CTL epitopes takes much time and large fund, many kinds of theoretical methods of predicting CTL epitopes come out, in which the Artificial Neural Networks is a most widely used one. In this paper, a model of predicting CTL epitopes, based on Artificial Neural Networks, is presented with two improvements of additional momentum term and adaptive learning rate. The results show that the modified model has some advanced properties compared with others, such as faster in processing and more accurate in predicting. With the training and the prediction of the epitope datum, veracities of the model as parameters are presented, as well as a set of optimized parameters. In comparison with existing results, predicting results given by our model are more satisfied.更多还原