基于Monte Carlo局部增强的多模态优化算法

Local Monte Carlo Search Approach to Multimodal Problem in Protein Conformation Space Optimization

高维构象空间搜索是蛋白质结构从头预测领域中一个亟需解决的关键问题。基于差分进化算法框架,提出了一种多模态蛋白构象空间优化算法。算法建立基于蛋白质空间特征向量的相似性测度指标,采用排挤更新策略,避免算法早熟,对蛋白质构象空间模态进行全局搜索;设计基于Monte Carlo局部搜索的片段组装方法,实现模态增强过程,有效平衡算法的收敛速度和种群多样性。采用Rosetta粗粒度能量模型,针对5种测试蛋白的实验结果表明:Monte Carlo局部增强和蛋白质特征向量的相似性测度能够有效地提高算法的性能,与Baker小组和Shehu小组的研究成果相比,提出的算法能够达到较高的预测精度,并得到一系列的亚稳态稳定结构。

We elucidated the native structure of a protein molecule from its sequence of amino acids.A problem known as de novo structure prediction,is a long standing challenge in molecular biology.High dimensional conformational space search is the key issue of protein structure prediction that is needed to be solved.Based on differential evolution algorithm framework,we proposed a multimodal protein conformational space optimization algorithm to address the multiple-minima problem in decoy sampling for de novo structure prediction.Algorithm builds the index of similarity measure that is based on the vectors of features of proteins,using exclusion strategy to implement global search.Local minimum search strategy with fragment assembly is able to avoid premature convergence,and can balance the convergence rate and the diversity of the population.A greedy search maps a child conformation to its nearest local minimum,and the molecular fragment replacement technique and differential evolution algorithm help child jump out of local minimum,thus the algorithm can get better search ability.Using Rosetta coarse-grained energy model,results show that the additional mini-mization and the exclusion strategy based on conformation space are key to obtaining a diverse ensemble of decoys.Compared with Baker research team and Shehu research team,the proposed algorithm can achieve better prediction accuracy.