预测蛋白质二级结构的人工神经元网络方法

本文独立地建立了用人工神经元网络预测蛋白质二级结构的方法,并通过分析我们提出的分布矩阵(表达每一类构象被预测成所有各类构象的可能性的矩阵),对于这一方法的误差以及造成误差的可能的原因进行了较过去更为深入的分析.并在此基础上提出了一种修正的学习方法,结果对于规则二级结构(α螺旋和β折叠)的预测精度和相关系数均有提高.

NEURAL NETWORK METHODS FOR PREDICTING THE SECODDARY STRUCTURE OF PROTEINS

We present the method for predicting the secondary structure of proteins usingartificial neural network models. By introducing and analyzing the Distributing Matrix which shows how each type of secondary structures is distributed among all the three types by the predicting aleorithm, the accuracy of the method and the possible reasons for mis-predicting are made clearer than before. Based on the analyzing a new modified learning algorithm is proposed which by using different training frequency for different type of secondary structure, gives better predicting accuracy and correlation coeffients for the regular secondary structures(alpha-helix and beta-sheet). We also noticed that the most frequently occurring mis-predictings are the ones between the regular secondary structures and the random coil. A possible explanation is given which states tha the error is partly due to the inaccuracy of secondary structure assignment.