基于副本交换的局部增强差分进化蛋白质结构从头预测方法

Replica Exchange Based Local Enhanced Differential Evolution Searching Method in Ab-initio Protein Structure Prediction

针对蛋白质高维构象空间搜索问题,提出一种基于副本交换的局部增强差分进化蛋白质结构从头预测方法(RLDE)。首先,采用基于知识的Rosetta粗粒度能量模型显著降低构象空间优化变量维数;其次,引入基于片段库知识的片段组装技术进一步减小构象搜索空间,有效避免搜索过程中的熵效应;此外,在每个副本层设置构象种群,采用差分进化算法对种群进行更新,然后利用Monte Carlo算法对种群做局部增强,以此得到全局和部分局部最优构象。综上,RLDE利用差分进化算法较强的全局搜索能力可以对构象空间进行有效的全局搜索;借助Monte Carlo算法局部搜索性能对构象空间局部极小区域进行更为充分的采样;副本交换策略保证了副本层中种群的多样性,同时能够增强算法跳出局部极小的能力,从而使得算法对构象空间的搜索能力进一步增强。15个目标蛋白测试结果表明,所提方法能够有效地对构象空间采样,得到高精度的近天然态蛋白质构象。

To address the searching problem in high-dimensional protein conformational space, a replica exchange based local enhanced differential evolution searching method in ab-initio protein structure prediction （RLDE） was proposed. In this paper, the knowledge-based coarse-grained energy model, Rosetta, was employed to considerably reduce the optimal variable dimension of protein conformational space,the knowledge-based fragment assembly technique was introduced to further reduce the dimension of protein conformational space. Thus the entropy effect caused by searching in high-di- mensionality conformational space could be avoided. Additionally, a conformation population was put into every replica layer, differential evolution algorithm was adopted to update the population in each layer, and the updated＂ populations were then enhanced by Monte Carlo method. As a consequence, the global optimal conformation and a series of metasta- ble conformations were generated. In conclusion, RLDE can effectively search the global conformational space through the strong global searching ability of differential evolution algorithm. The well local searching performance of Monte Carlo is also employed to sample the local minimum area adequately. Replica exchange strategy ensures the diversity of population in replica layers, and the capacity of algorithm to jump out of local minimum is enhanced as well, thereby makes the searching ability further heightened. Test results of 15 target proteins show that the proposed method can generated high-resolution near-native protein conformations by searching the conformational space effectively.