Learning Metrics for Visualizing Gene Functional Similarities

The usual first step in analyzing the large and high-dimensional data sets measured by microarrays is visual exploration. In this work self-organizing maps have been used to visualize similarity relationships of data samples. In all unsupervised data analysis methods the measure of similarity determines the result; we propose to use the learning metrics principle to derive a metric from interrelationships between data sets. A metric is derived for a gene knock-out expression data set by considering those changes in the expression space that cause changes in the functional classes of the genes to be more important. The genes for which the new metric is the most different from the usual correlation metric are listed and visualized with a self-organizing map computed in the new metric.