基于PSO_BP神经网络的固有无序蛋白质预测  

Prediction of Intrinsically Disordered Proteins Based on PSO_BP Neural Network

作　　者：杨佳 刘海员 Yang Jia;Liu Haiyuan(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350

出　　处：《南开大学学报（自然科学版）》2023年第1期1-7,共7页Acta Scientiarum Naturalium Universitatis Nankaiensis

摘　　要：通过对蛋白质有序-无序区域分析,找到表示氨基酸的34维特征.其中,样本熵是用于计算时间序列复杂度的参数,通过比较20种氨基酸在两种区域的出现频率,将其对应为0-9的时间序列,从而计算蛋白质的复杂度.另外,使用长度为35的滑动窗口将相邻氨基酸联系起来,提高了预测准确度.最后,使用粒子群算法优化BP神经网络的节点参数,训练并实现有序-无序分类的5个网络,取均值后转化为有序-无序输出.使用DisProt数据集和R80数据集分别进行十折交叉验证,预测准确率分别达76%和87%以上.By analyzing the ordered-disordered regions of proteins, the 34-dimensional features that represent amino acids were found. Particularly, the sample entropy is a parameter used to calculate the complexity of time series. By comparing the appearance frequency of 20 amino acids in the two regions, it corresponds to a time series of 0-9 to calculate the complexity of the protein. In addition, a 35-length sliding window is used to connect adjacent amino acids, which greatly improved the prediction accuracy. Finally, the PSO algorithm is used to optimize the parameters of the BP neural network, and 5 nets were trained, the average value can be converted into ordered-disordered output. Using Dis Prot data set and R80 data set to perform ten-fold crossvalidation, the prediction accuracy rates reached to 76% and 87% respectively.

关 键 词：生物信息学 固有无序蛋白质 神经网络 粒子群优化算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]