Description
Interactive Visual Analysis of Mass Cytometry Data by Hierarchical Stochastic Neighbor Embedding Reveals Rare Cell Types
Thomas Höllt1,2, Vincent van Unen3, Nicola Pezzotti1, Na Li3, Marcel Reinders4, Elmar Eisemann1, Frits Koning3, Anna Vilanova1, Boudewijn Lelieveldt4,5
1Computer Graphics and Visualization Group, TU Delft, the Netherlands; 2Computational Biology Center, Leiden University Medical Center, the Netherlands; 3Department of Immunohematology and Blood Transfusion, Leiden University Medical Center, the Netherlands
4Pattern Recognition and Bioinformatics Group, TU Delft, the Netherlands; 5Division of Image Processing, Department of Radiology, Leiden University Medical Center, the Netherlands
Mass cytometry allows high-resolution dissection of the cellular composition of the immune system. However, the high-dimensionality, large size, and non-linear structure of the data poses considerable challenges for data analysis.
We introduce Hierarchical Stochastic Neighbor Embedding (HSNE) for single-cell analysis, a computational approach that constructs a hierarchy of non-linear similarities, allowing the analysis of millions of cells via different levels of detail up to single-cell resolution within minutes. We integrated HSNE into the Cytosplore+HSNE framework to facilitate interactive exploration and analysis of the hierarchy by a set of corresponding two-dimensional plots with stepwise increase in detail up to the single-cell level. This divide and conquer approach minimizes computation time and, thereby, allows efficient and interactive visualization.
We validated the discovery potential of Cytosplore+HSNE by re-analyzing a recent study on gastrointestinal disorders as well as two other publicly available mass cytometry datasets. We found that Cytosplore+HSNE efficiently identifies both abundant and rare cell populations, without resorting to downsampling of the data, including rare cell populations that were missed in a previous analysis due to downsampling. Taken together, Cytosplore+HSNE offers unprecedented possibilities for visual exploration and analysis of millions of cells measured in mass cytometry studies.
This work was supported by the STW Project 12720.