氨基酸邻域对二级结构的影响程度研究

通过研究神经网络权值矩阵的算法,挖掘蛋白质二级结构与氨基酸序列间的内在规律,提高一级序列预测二级结构的准确度。神经网络方法在特征分类方面具有良好表现,经过学习训练后的神经元连接权值矩阵包含样本的内在特征和规律。研究使用神经网络权值矩阵打分预测;采用错位比对方法寻找敏感的氨基酸邻域;分析测试集在不同加窗长度下的共性表现。实验表明,在滑动窗口长度L=7时,预测性能变化显著;邻域位置P=4的氨基酸残基对预测性能有加强作用。该研究方法为基于局部序列特征的蛋白质二级结构预测提供了新的算法设计。

Neural Network Matrix Studies of Protein Secondary Structures Determined by the Interaction of Local Amino Acids

Through the studying of algorithms for weight matrix of neural networks and the instinct regularity dining between secondary structure of proteins and amino acid sequences,we aim to improve the accuracy of prediction from primary structure to secondary structure.Neural network methodology proves to be working well in characterization and prediction of complex system,and the weight matrix contains the instinct information of examples after training.This study employing neural network matrix score prediction method,developed the more sensitive ways locating the critical neighboring amino acids through offsets or mismatch comparisons;different length of loading windows of the amino acids were characterized in the testing assets.Experiments shows that the prediction capability were significantly fluctuated when sliding window length L=7.Further studies indicated that when the fourth neighboring amino acid（P=4） was incorporated into the loading windows,the results have demonstrated more enhanced accuracy than those predictions that did not incorporate the 4th neighboring amino acid.The research method provides new design of algorithm for the prediction of protein secondary structure based on local sequence characteristic.