Data visualization and structure identification

Abstract

For three datasets, all dealing with materials with ABO3 chemistries, the two data visualizations algorithms of Tsafrir et al. [Bioinformatics 21, 2301 (2005)] were studied and applied. These algorithms permute the distance matrix associated with the data in a way to unveil structure in one case by keeping large-distanced information afar or in the other case by keeping small-distanced information near. Modifications to their proposed numerical implementations were made to enhance effectiveness. The two algorithms were used both in space of the materials and the features, looking for groupings of features and materials. In general, for the datasets considered, when visualized, the features tended to show more distinctive structure (clustering) than the materials. For enhanced grouping of materials, the initial studies point to the importance of feature selection.