College of Computer Science, Shaanxi Normal University, Xi'an, P R China.
Proteomics. 2013 Jan;13(2):278-90. doi: 10.1002/pmic.201200309.
As is known to all, traditional clustering algorithms do not work well due to the topological features of protein-protein interaction networks. An improved clustering method based on bacteria foraging optimization (BFO) mechanism and intuitionistic fuzzy set, short for improved BFO, is proposed in this paper, in which the trigonometric function is used to define the membership degrees and the indeterminacy degree is introduced to detect the overlapping modules. In chemotactic operation of BFO, the algorithm initializes a cluster center according to comprehensive network feature value of node and eliminates the isolated point in accordance with edge-clustering coefficient. In the reproduction operation of BFO, the nodes possessing high membership degrees are merged into the cluster that the cluster center belongs to and labeled as visited nodes. Meanwhile, the nodes that also have high indeterminacy degrees are visited again when generating another cluster. The procedure of elimination-dispersal operation is equivalent to the selection of the next cluster center. Finally, the algorithm merges the clusters having high similarity. The results show that the algorithm not only determines the cluster number automatically, improves the f-measure value of cluster results, but also identify the overlaps in protein-protein interaction network successfully.
众所周知，由于蛋白质-蛋白质相互作用网络的拓扑特征，传统的聚类算法效果不佳。本文提出了一种基于细菌觅食优化（BFO）机制和直觉模糊集的改进聚类方法，简称改进 BFO。在该方法中，使用三角函数定义隶属度，引入不确定度来检测重叠模块。在 BFO 的趋化操作中，该算法根据节点的综合网络特征值初始化聚类中心，并根据边聚类系数消除孤立点。在 BFO 的繁殖操作中，具有高隶属度的节点被合并到聚类中心所属的聚类中，并标记为已访问节点。同时，在生成另一个聚类时，会再次访问具有高不确定度的节点。消除-扩散操作的过程相当于选择下一个聚类中心。最后，该算法合并具有高相似度的聚类。结果表明，该算法不仅可以自动确定聚类数，提高聚类结果的 F 值，而且还可以成功识别蛋白质-蛋白质相互作用网络中的重叠。
