Multidimensional scaling for genomic data

Abstract

Scientists working with genomic data face challenges to analyze and understand an ever-increasing amount of data. Multidimensional scaling (MDS) refers to the representation of high dimensional data in a low dimensional space that preserves the similarities between data points. Metric MDS algorithms aim to embed inter-point distances as close as the input dissimilarities. The computational complexity of most metric MDS methods is over O(n2), which restricts application to large genomic data (n > 106). The application of non-metric MDS might be considered, in which inter-point distances are embedded considering only the relative order of the input dissimilarities. A non-metric MDS method has lower complexity compared to a metric MDS, although it does not preserve the true relationships. However, if the input dissimilarities are unreliable, too difficult to measure or simply unavailable, a non-metric MDS is the appropriate algorithm. In this paper, we give overview of both metric and non-metric MDS methods and their application to genomic data analyses.