Solving 2D HP Protein Folding Problem by Parallel Ant Colonies

Abstract

To predict protein structure based on HydrophobicPolar model(HP model) in two-dimensional space is called 2D HP protein folding problem. Ant Colony Optimization(ACO), which is inspired by the foraging behavior of ants, is a popular heuristic approach for solving combinatorial optimization problems. This paper presents a method of solving the 2D HP protein folding problem by parallel ACO algorithm. Each ant colony is able to search the best solution guided by the shared pheromone matrix which accumulates the good experience achieved by previous populations. The shared pheromone matrix can integrate all the search knowledge found by parallel colonies. Experimental results show that the parallel implementation performs better comparing with the other ACO solutions.

Parallel metaheuristics: recent advances and new trends

“…Among the most important ones in recent years, we can mention the following: Automation and robotics, where accurate and efficient learning algorithms to control and use the information technologies are required to reduce the need for human presence. Here, parallel metaheuristics provide a decisive help to tackle learning problems that handle large volume of data (Bouamama, ), and those control problems that involve complex training procedures (Hereford, , ; Huang et al., ). Bioinformatics, an emergent scientific field where parallel models of metaheuristics are helpful tools to cope with computationally expensive optimization problems in molecular biology that often also need to manage very large amount of data, such as sequence alignment (Gomes et al., ; Zola et al., ), DNA sequencing (Hongwei and Yanhua, ; Wirawan et al., ), gene finding (Rausch et al., ), genome assembly (Alba and Luque, ; Nebro et al., ), drug design (Boisson et al., ), protein structure alignment/prediction (Chu and Zomaya, ; Guo et al., ; Islam and Ngom, ; Tantar et al., ), phylogenetic inference (Blagojevic et al., ; Cancino et al., ; Grouchy et al., ), and other related problems (Guarracino et al., ; Martins et al., ; Nebro et al., ). Engineering design, where systems have many components, a large design space, and they usually involve functions with huge computation demands. These characteristics make parallel metaheuristics one of the most promising alternatives to get accurate solutions in reasonable execution times for complex tasks such as aerodynamic optimization and airfoil design (Asouti and Giannakoglou, ; Lim et al., ), design optimization of turbomachinery blade rows (Sasaki et al., ), electronic circuit and VLSI design (Alba et al., ; Lau et al., ; Sait et al., , ), antenna design (Kalinli et al., ; Weis and Lewis, ), signal processing (Li et al., ), etc. Hydraulic engineering, where parallel metaheuristics have been used to efficiently deal with real‐world scenarios arising in water supply network design optimization (López‐Ibánez, ), groundwater source identification (Babbar and Minsker, ; Mirghania et al., ; Sinha and Minsker, ), and multiobjective groundwater problems (Tang et al., ). Information processing, classification, and data mining, where parallel metaheuristics significantly he...…”

“…Some parallel metaheuristics with recent implementations of the fine‐grain model in multicore architectures includes EAs (Munawar et al., ), EDAs (Perez et al., ), MOEAs (Nebro and Durillo, ), ACOs (Tsutsui and Fujimoto, ), VND/ILS (Subramanian et al., ), TS and several other metaheuristics (Bozejko et al., ). The master–slave model for parallel metaheuristics has also been implemented in multicore processors, for ACO (Delisle et al., , ; Guo et al., ; López‐Ibánez, ; Tsutsui, ; Tsutsui and Fujimoto, ), EA (Cardenas et al., ), TS and branch and bound (Hung and Chen, ); in these methods, the main advantage is the ability of computing the fitness evaluation in parallel by using several threads. Multicore multipopulation methods have also been proposed for several metaheuristics, such as ACO (Delisle et al., ; Gao et al., ; Li et al., ; Lucka and Piecka, ; Xiong et al., , ), EAs (Byun et al., ; He et al., ; Tsutsui, ), PSO (Tu and Liang, ), and the parallel artificial bee colony algorithm (Narasimhan, ).…”

从头预测蛋白质骨架的一种并行蚁群方法及其在casp8/9 中的应用

“…通过这个共享信息素矩阵, 每个蚁群可以同步地交 换搜索经验, 共同朝着更好的目标值演化. 共享信息矩阵策略已经成功地应用到诸如 TSP [22] 、Bayes 网络学习 [23] 、蛋白质 HP 折叠问题 [24] 和 QAP [25] 上. Rosetta3.x 中的聚类协议 [19] 基于结构 RMSD (root mean square deviation) 相似度, I-TASSER 的 聚类协议 SPICKER [26] 采用 TM-score (template modeling score) 进行聚类. 由于聚类方法各有优缺 点, 结构相似性尺度的选择也会与能量函数表现自由能的程度有关 [27] , 所以针对不同的预测目标, 本 文采用了两种不同的聚类方法.…”

“…24 日发布了第一个预测目标, 每个工作日发布 3 个目标, 每个目标可以 有 72 小时的计算时间. pacBackbone 以 LenServer 为名参加了 Server 组的比赛, 对所有 118 个预测 目标共提交 588 个结果.…”

A novel state space representation for the solution of 2D-HP protein folding problem using reinforcement learning methods

“…Thus, in this study the reinforcement learning methods are used for the solution of the protein-folding problem in two dimensional lattice model. There exist many studies [4][5][6][7][8][9] in literature that proposed different methods for the solution of this problem, but the use of reinforcement learning methods are quite new. In [10][11][12][13], authors used the Q-learning algorithm to solve the protein folding problem in two dimensional hydrophobic-polar (2D-HP) model.…”