Clustering and dimensionality reduction Clustering and clustering algorithms

Clustering is the method of establishing structure within a collection of unlabeled data. Clusters are organized according to points that are similar to each other as well as different from points in other clusters. Measuring similarity entails either plotting gene expression of each experiment or measuring correlation under different parametric (Pearson or Euclidean) or non-parametric (Spearman or Kendall) conditions. The main goal of the clustering algorithm is to, therefore, create clusters that are internally coherent while simultaneously different from each other. Commonly used clustering analysis methods include hierarchical clustering and flat clustering. Hierarchical clustering creates an informative hierarchy of clusters whereas flat clustering creates a flat group of clusters that does not possess any real structure that would relate clusters to one another. Hierarchical clustering consists of algorithms, such as single-link, complete-link, group-average, and centroid similarity, each of which differ in their similarity measures. In contrast, flat clustering comprises algorithms such as K-means in which the main purpose is to minimize the average squared Euclidean distance of certain points from their cluster centers.