动态步长蛋白质构象空间搜索方法  

Ab-initio dynamic-step-size searching of protein conformational space

作　　者：张贵军[1] 郝小虎[1] 周晓根 秦传庆 

机构地区：[1]浙江工业大学信息工程学院,杭州310023

出　　处：《吉林大学学报（工学版）》2016年第2期585-594,共10页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(61075062;61573317);浙江省自然科学基金项目(LY13F030008);浙江省科技厅公益项目(2014C33088);浙江省重中之重学科开放基金项目(20120811);杭州市产学研合作项目(20131631E31)

摘　　要：针对蛋白质构象空间采样问题,提出了一种基于能量引导树搜索框架的动态步长构象空间搜索方法。通过蛋白质构象特征提取,将高维二面角优化空间映射到低维结构特征向量空间,有效避免了维数灾难问题;根据能量和温度测度离散化特征空间为多个能量层和温度层,并系统划分为'构象室',减小构象空间搜索范围。在不同能量层,赋予相应的片段组装步长和蒙特卡洛扰动步长,在不同温度层,采用相应Metropolis准则接收当前构象;辅以副本交换方法,增强对构象空间中稳态结构的采样能力。12个蛋白质测试结果表明,该方法可以快速有效地采样得到近天然态构象。To address the sampling problem of protein conformational space,an Ab-initio dynamic-stepsize searching method of protein conformational space is proposed.This method is based on the energy tree-based searching framework. The high-dimensional optimization space of dihedral angle is projected to a low-dimensional space of feature vector with the protein conformation feature extraction,effectively avoiding the curse of dimensionality problem.The feature space is discretized according to the energy and temperature.Then the layers are systematically divided into cells to reduce the searching space.Relevant Fragment Assembly（FA）step-size and Monte Carlo disturbance step-size are set according to the specific energy layer,and the corresponding Metropolis criterion is employed to accept the conformation within different temperature layers.The replica-exchange method is used as auxiliary method to enhance the sampling of native-like protein conformation.Test results of 12 proteins show that their native-like protein conformations can be reached successfully and effectively by the proposed method.

关 键 词：人工智能 从头预测 树搜索 动态步长 片段组装 蒙特卡洛 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]