Thumbnails:
List:
Year:
Category:
Session:
Poster:
Getting poster data...
Karel Drbal (Faculty of Science, Charles University, Prague, Czechia)
Systematic, easily intelligible depiction of changes in large, high-dimensional datasets creates a complicated challenge for data visualization algorithms. We present an algorithm that overcomes the currently available methods by combining highly effective clustering with rigorous statistical testing and dimensionality reduction. The algorithm (called DiffSOM) uses self-organizing maps both as a basis for the 2-dimensional EmbedSOM display, and for high-precision clustering of the dataset that is used for statistical testing. In result, a fast, accurate and comprehensible color-coded overview of significant changes in large multidimensional datasets is provided. We demonstrate the process by visualizing changes in data from a recent immunology study, which examined differences in mass-cytometry samples of the peripheral blood cells from pregnant women taken at different time points and with various experimental in vitro perturbations. Linear computational complexity of the approach additionally guarantees quick processing of large amounts of data - 1 million 36-dimensional data points from 18 individuals were embedded and tested in less than 5 minutes on a laptop computer.