Hierarchical Cluster Analysis (HCA) of Microorganisms: An Assessment of Algorithms for Resonance Raman Spectra
Hierarchical Cluster Analysis (HCA) of Microorganisms: An Assessment of Algorithms for Resonance Raman Spectra
复制标题
DOI：
10.1366/10-06064
复制
发表时间：
2011-02-01
影响因子：
3.5
通讯作者：
Meinhardt-Wollweber, Merve
中科院分区：
文献类型：
--
作者：
Kniggendorf, Ann-Kathrin;Gaul, Tobias William;Meinhardt-Wollweber, Merve
共振拉曼显微光谱与层次聚类分析 (HCA) 相结合是快速检查复杂生物和医学样品的最有前途的工具之一。 HCA 是一种现成的计算机化工具，用于检查大量数据的共同特征，多年来已经开发了多种用于此目的的算法。然而，从复杂生物样品获得的共振拉曼光谱可能源自不同的发色团，也可能源自不同宿主环境（即细菌）中发现的共同发色团。因此，应用于共振拉曼光谱的算法必须在相同的无监督分析中处理具有高度内在相似性的数据，即源自共同发色团的光谱，以及具有高度不同特征的数据，即来自不同发色团的光谱。我们检查了八种广泛使用的用于细菌共振拉曼光谱聚类分析的层次聚类分析算法的性能：单链接（最近邻）、完全链接（最远邻）、平均链接、加权平均链接、质心、中值和 Ward 算法。通过将一组高质量参考光谱的聚类结果与一组从单细胞记录的光谱聚类时获得的结果进行比较来评估算法性能。通过对单个细胞的 100 个光谱进行平均来形成参考。虽然所有算法在对参考光谱进行聚类时都返回高度相似的结果，但在应用于单个光谱时，它们的性能存在显着差异。性能最佳的算法加权平均关联正确地对单个光谱进行了分组，可靠性高于 95%，而簇之间的光谱距离与参考光谱获得的结果偏差不到 10%。相反，表现最差的算法与参考聚类根本没有相似之处。广泛使用的 Ward 算法在光谱距离上与参考值的偏差高达 30%，并返回表达相同发色团的细菌之间不同的光谱关系。
Resonance Raman microspectroscopy in combination with hierarchical cluster analysis (HCA) is one of the most promising tools for the rapid examination of complex biological and medical samples. HCA is a ready, computerized tool for examining large sets of data for common characteristics, and a multitude of algorithms for this purpose have been developed over the years. However, resonance Raman spectra obtained from complex biological samples may originate from different chromophores as well as from a common chromophore found in different host environments, i.e., bacteria. Therefore, algorithms applied to resonance Raman spectra must handle data of high intrinsic similarity, i.e., spectra originating from a common chromophore, and data with highly dissimilar features, i.e., spectra from different chromophores, in the same unsupervised analysis. We examined the performance of eight widely used algorithms for hierarchical cluster analysis in clustering resonance Raman spectra of bacteria: Single-Linkage (Nearest-Neighbor), Complete-Linkage (Farthest-Neighbor), Average-Linkage, Weighted-Average-Linkage, Centroid, Median, and the Ward algorithm. Algorithm performance was evaluated by comparing the results of clustering a set of high-quality reference spectra with the results obtained when clustering a set of spectra recorded from single cells. References were formed by averaging 100 spectra of individual cells. While all algorithms returned highly similar results when clustering the reference spectra, their performance differed significantly when applied to single spectra. The best-performing algorithm, Weighted-Average-Linkage, correctly grouped single spectra with a reliability of above 95% while the spectral distances between the clusters deviated less than 10% from the results obtained with reference spectra. In contrast, the algorithm performing worst showed no similarity to the reference clustering at all. The widely used Ward algorithm deviated up to 30% from the reference in the spectral distances and returned a different spectral relation between bacteria expressing the same chromophore.