Two Dimensional Sort Regions For Multi-Dimensional Cell Clusters
F. L. Battye
The Walter & Eliza Hall Institute
Flow Cytometry Cluster Analysis techniques explicitly define the positions of sub-populations of cells within multi-dimensional data spaces. While techniques for sorting each of the sub-populations so defined are under development (pattern sorting, neural networks), most cell sorters are still limited to sort criteria based on logical combinations of "regions" defined on 2D projections of the data (dot plots, contour maps). However, in many cases, there is no 2D projection for which the wanted cluster is clearly separable and the establishment of sort regions requires an iterative, trial and error process. Even then, it is clearly not possible to perfectly encircle a sub-population defined in a data space of 6 or 8 dimensions by combining 2D regions
Hence, software has been developed for automating this process. The program firstly selects the best set of parameter pairs for 2D projections of the data. The choice is based on a measure of "exposure" of the required cells, i.e. the proportion of these cells not overlapped by other populations. Next, within these projections, regions are constructed which are then combined into a sort gate for the given cell sub-population. The regions begin with boundaries in each projection which enclose a large proportion (~98%) of the required cells. They are then iteratively whittled back by removing edges at which the local purity is lowest. As expected, the extent to which these sort regions need to be whittled back depends on a trade-off between purity and recovery and the end point must be determined by the human operator.