基于FCM算法的小脑基因模糊聚类分析

探讨基因表达数据的聚类分析方法,结合一种聚类结果的评判准则,应用于胎儿小脑基因表达数据,得到了最优的聚类结果,并做出了生物学解释.利用Matlab软件进行了仿真,利用模糊聚类Xie-Beni指数得到了最优聚类数,并把每一类对应的基因标号输出到txt文件,最后进行生物学解释.得到的小脑基因最优聚类数为3类,与生物学意义比较吻合,各类中的基因功能接近.基于FCM算法的基因模糊聚类是有效的,结果具有一定生物学意义,能对生物学基因聚类有一定指导作用.     

一个新的模糊聚类有效性指标

提出了一个判别模糊聚类中聚类数有效性的新指标.首先利用FCM算法对数据集进行模糊聚类,通过隶属度矩阵和聚类中心构建加权二分网络.然后通过改进加权二分网络的模函数,定义一个新的聚类有效性指标.为了检验该有效性指标的性能,选取了三个常见的有效性指标在十五个数据集上进行了对比.实验结果表明,该有效性指标具有较好的性能.     

开源设计网络社区的用户聚类研究

为了对开源设计网络社区中的用户进行细分研究,首先采用复杂社会网络技术建立了社区组织的复杂网络模型.其次,根据用户在社区中的三种行为设立了备选指标,并通过指标聚类的方法对指标进行筛选,保证了聚类指标的全面性和代表性.在此基础上,以OpenIDEO为研究案例,采用K-Means算法对该社区中的用户进行了聚类,根据聚类结果将用户分为了创新型用户、传播型用户和普通用户,并进一步分析了各类用户的特点及参与动机.聚类结果表明,所提出的聚类指标及算法可以清晰地识别出开源社区的用户类型及占比,为开源设计社区管理机制和激励措施的优化提供了准确的依据.     

混合优化RBF-BP网络的板形缺陷识别

针对传统板形模式识别方法存在精度低、鲁棒性弱的问题,提出了一种混合优化RBF-BP组合神经网络板形模式识别方法。首先利用自组织映射网络(SOM)对样本聚类,利用聚类后的网络拓扑结构确定RBF的中心,并计算RBF的宽度,克服了传统聚类算法随机选取中心导致聚类结果不稳定的问题。然后利用遗传算法(GA)良好的全局搜索能力优化整个网络的权值。RBF-BP组合神经网络是由一个RBF子网和一BP子网串联构成的,该网络同时具备BP神经网络能较好地预测未知样本的能力以及RBF神经网络的逼近速度快的优点。并以某900HC可逆冷轧机板形识别为应用背景,在MATLAB2010a环境下进行仿真实验,结果表明混合优化RBF-BP组合神经网络的板形模式识别方法能够识别出常见的板形缺陷,提高了板形缺陷识别精度并具有较好的鲁棒性,可以满足板带轧机高精度的板形控制要求。     

基于ICA的时间序列聚类方法及其股票数据分析中的应用

时间序列聚类分析是时间序列数据挖掘中的重要任务之一,通常由于时间序列数据的特殊结构,导致一般的聚类算法不能直接应用于时间序列数据.本文提出了一种基于独立成分分析与改进K-均值算法相结合的时间序列聚类算法,该算法首先利用独立成分分析对时间序列数据进行特征提取,然后利用改进K-均值聚类算法完成对时间序列特征数据的聚类分析,从而得到了一种新的基于特征的时间序列聚类方法.为了验证该方法的有效性和可行性,将其应用于实际的股票时间序列数据聚类分析中,取得了较好的数值结果.     

基于Copula函数的非线性时间序列聚类

时间序列数据的聚类是对面板数据或多维时间序列根据序列相似度进行分组。聚在同一组的时间序列具有相近的模型参数,尤其是当序列较短时聚类后能够得到更精确的参数估计。现存的时间序列聚类方法的距离度量大都基于时间序列的线性假设,但是现实中时间序列通常是非线性的。本文提出了一种基于Copula距离测度的非线性时间序列数据的聚类方法,它利用了Copula函数获取时间序列的非线性相依结构。作为一种非参数的距离度量,基于Copula函数的距离度量能够识别动态相关结构的相似性。大量的模拟实验和实证研究验证了我们所提方法的有效性。     

基于学科布局的科研机构层次聚类模型构建与实证分析

研究不同机构学科布局或主题领域分布的差异,并基于学科布局对科研机构进行聚类分析,有助于掌握科研机构的发展规律,对提高机构管理的效率和效益方面有着重要意义.文章采用科学基金项目数据来衡量科研机构学科布局情况,并针对基于项目数据学科布局聚类过程中数据存在的高维、稀疏特征,传统的聚类方法难以直接处理的问题,构建了"t-SNE+层次聚类"集成聚类方法.通过对中国科学院的国家自然科学基金项目数据进行研究,得到了中国科学院下属的117个研究机构的基于学科布局的聚类结果,并以10类为例,阐述了不同类别的机构学科布局特征.实证结果表明,文章使用的"t-SNE+层次聚类"方法得到的聚类结果,相比于传统方法得到的结果具有更好的效果,可以为学科布局调整提供支撑.     

一种基于核密度聚类的电力负荷模式精细化识别算法研究

密度峰值聚类算法(DPC)是一种基于密度的非监督学习算法.分析用电类型复杂的电力负荷数据集时,存在负荷曲线聚类效果过分依赖人为参数设定和无法识别潜在用电模式的缺陷.结合非参数核密度估计,使用带宽搜索与边界优化提出一种适应多类型复杂用户的电力负荷数据优化聚类算法.在某市10KV真实数据集中进行算法测试,使用Davies-Bouldin有效性指标对比优化前后算法聚类效果.结果表明改进算法在面向用户类型复杂的电力数据集时,能够实现已知用电模式精确识别与潜在用电模式的深度挖掘并显著提高聚类有效性.     

利用提升小波的蛋白质相互作用特征提取

利用提升小波从蛋白质序列中提取出它们相互作用的频谱特征,经支持向量机训练学习后,用于预测蛋白质间的相互作用.模拟计算结果表明,在阳性数据和阴性数据平衡的前提下,利用提升小波获取的低维蛋白质相互作用特征向量可以得到较高预测精度.进一步阐述了不同物种的蛋白质相互作用网络有着不同特征,为了得到更准确的预测结果,需要利用不同的方法提取蛋白质相互作用的特征.     

融合系统聚类与BP神经网络的世界碳排放预测模型研究

针对碳排放指标的复杂性与多样性,利用系统聚类对世界碳排放指标进行筛选;然后运用BP神经网络对世界碳排放量进行预测.结果表明运用系统聚类方法分析碳排放指标,简化了BP神经网络输入层,使网络训练得到较快实现,相比传统BP神经网络预测方法具有更高的精度,为碳排放预测及其它相关预测提供了一种新的、可供借鉴的方法.     

Integration of topological measures for eliminating non-specific interactions in protein interaction networks

High-throughput protein interaction assays aim to provide a comprehensive list of interactions that govern the biological processes in a cell. These large-scale sets of interactions, represented as protein–protein interaction networks, are often analyzed by computational methods for detailed biological interpretation. However, as a result of the tradeoff between speed and accuracy, the interactions reported by high-throughput techniques occasionally include non-specific (i.e., false-positive) interactions. Unfortunately, many computational methods are sensitive to noise in protein interaction networks; and therefore they are not able to make biologically accurate inferences.In this article, we propose a novel technique based on integration of topological measures for removing non-specific interactions in a large-scale protein–protein interaction network. After transforming a given protein interaction network using line graph transformation, we compute clustering coefficient and betweenness centrality measures for all the edges in the network. Motivated by the modular organization of specific protein interactions in a cell, we remove edges with low clustering coefficient and high betweenness centrality values. We also utilize confidence estimates that are provided by probabilistic interaction prediction techniques. We validate our proposed method by comparing the results of a molecular complex detection algorithm (MCODE) to a ground truth set of known Saccharomyces cerevisiae complexes in the MIPS complex catalogue database. Our results show that, by removing false-positive interactions in the S. cerevisiae network, we can significantly increase the biological accuracy of the complexes reported by MCODE.     

关于DNA序列分类问题的模型

本文提出了一种将人工神经元网络用于 DNA分类的方法 .作者首先应用概率统计的方法对 2 0个已知类别的人工 DNA序列进行特征提取 ,形成 DNA序列的特征向量 ,并将之作为样本输入 BP神经网络进行学习 .作者应用了 MATLAB软件包中的 Neural Network Toolbox(神经网络工具箱 )中的反向传播 ( Backpropagation BP)算法来训练神经网络 .在本文中 ,作者构造了两个三层 BP神经网络 ,将提取的 DNA特征向量集作为样本分别输入这两个网络进行学习 .通过训练后 ,将 2 0个未分类的人工序列样本和 1 82个自然序列样本提取特征形成特征向量并输入两个网络进行分类 .结果表明 :本文中提出的分类方法能够以很高的正确率和精度对 DNA序列进行分类 ,将人工神经元网络用于 DNA序列分类是完全可行的     

一个基于复制变异准则的蛋白质作用网络的构建模型

本文主要根据生物学的复制和变异的基本原理,提出了具有反偏爱复制特性的一个蛋白质作用网络的新的演化模型.通过计算,发现所得的网络与实测的蛋白质作用网络的某些拓扑性质能够很好的吻合.生成的网络不仅是稀疏的,而且具有小世界性和无标度(scale-free)性质.     

A Novel Mixed Integer Linear Programming Model for Clustering Relational Networks

Integer programming models for clustering have applications in diverse fields addressing many problems such as market segmentation and location of facilities. Integer programming models are flexible in expressing objectives subject to some special constraints of the clustering problem. They are also important for guiding clustering algorithms that are capable of handling high-dimensional data. Here, we present a novel mixed integer linear programming model especially for clustering relational networks, which have important applications in social sciences and bioinformatics. Our model is applied to several social network data sets to demonstrate its ability to detect natural network structures.     

Scientific collaboration network of Turkey

Networking via co-authorship is an important area of research and used in many fields such as ranking of the universities/departments. Studying on the data supplied by the Web of Science, we constructed a structural database that defines the scientific collaboration network of the authors from Turkey, based on the publications between 1980 and 2010. To uncover the evolution and structure of this complex network by scientific means, we executed some empirical measurements. The Turkish scientific collaboration network is in an accelerating phase in growth, highly governed by the national policies aiming to develop a competitive higher education system in Turkey. As our results suggest the authors tend to make more number of collaborations in their studies over the years. The results also showed that, node separation of the network slightly converges about 4, consistent with the small world phenomenon. Together with this key indicator, the high clustering coefficient, (which is about 0.75) reveals that our network is strongly interconnected. Another quantity of major interest about such networks is, “the degree distribution”. It has a power-law tail that defines the network as scale-free. Along with the final values, the time evolutions of the above-mentioned parameters are presented in detail with this work. In a good agreement with the recent studies, our network yields some significant differences especially in growing rate, clustering properties and node separation. In contrast with the recent studies, we also showed that preferring to attach popular nodes result with being a more popular node in the future.     

General results on preferential attachment and clustering coefficient

This is a review paper that covers some recent results on the behavior of the clustering coefficient in preferential attachment networks and scale-free networks in general. The paper focuses on general approaches to network science. In other words, instead of discussing different fully specified random graph models, we describe some generic results which hold for classes of models. Namely, we first discuss a generalized class of preferential attachment models which includes many classical models. It turns out that some properties can be analyzed for the whole class without specifying the model. Such properties are the degree distribution and the global and average local clustering coefficients. Finally, we discuss some surprising results on the behavior of the global clustering coefficient in scale-free networks. Here we do not assume any underlying model.     

Common cold outbreaks: A network theory approach

In this study, at first we evaluated the network structure in social encounters by which respiratory diseases can spread. We considered common-cold and recorded a sample of human population and actual encounters between them. Our results show that the database structure presents a great value of clustering. In the second step, we evaluated dynamics of disease spread with SIR model by assigning a function to each node of the structural network. The rate of disease spread in networks was observed to be inversely correlated with characteristic path length. Therefore, the shortcuts have a significant role in increasing spread rate. We conclude that the dynamics of social encounters’ network stands between the random and the lattice in network spectrum. Although in this study we considered the period of common-cold disease for network dynamics, it seems that similar approaches may be useful for other airborne diseases such as SARS.     

The clustering coefficient and the diameter of small-world networks

The small-world network, proposed by Watts and Strogatz, has been extensively studied for the past over ten years. In this paper, a generalized small-world network is proposed, which extends several small-world network models. Furthermore, some properties of a special type of generalized small-world network with given expectation of edge numbers have been investigated, such as the degree distribution and the isoperimetric number. These results are used to present a lower and an upper bounds for the clustering coefficient and the diameter of the given edge number expectation generalized small-world network, respectively. In other words, we prove mathematically that the given edge number expectation generalized small-world network possesses large clustering coefficient and small diameter.     

A social network model with proximity prestige property

In general, many real-world networks not only possess scale-free and high clustering coefficient properties, but also have a fast information transmission capability. However, the existing network models are unable to well present the intrinsic fast information transmission feature. The initial infected nodes and the network topology are two factors that affect the information transmission capability. By using preferential attachment to high proximity prestige nodes and triad formation, we provide a proximity prestige network model, which has scale-free property and high clustering coefficient. Simulation results further indicate that the new model also possesses tunable information transmission capability archived by adjusting its parameters. Moreover, comparing with the BA scale-free network, the proximity prestige network PPNet05 achieves a higher transmission capability when messages travel based on SIR and SIS models. Our conclusions are directed to possible applications in rumor or information spreading mechanisms.