High-Performance Visualization of Multi-Dimensional Gene Expression Data

Previous application of Kohonen's self organizing map to common visualizations has yielded promising results. In this research, we extend the classic two-dimensional scatter plot visualization algorithm into the third dimension by permitting competition to occur within a three-dimensional search space. This approach takes advantage of spatial memory and increases the intrinsic dimensionality of a widely used visualization technique. We also present a method of parallelizing this novel algorithm as a method of overcoming the runtime complexity associated with it using MPI. We note that this algorithm responds extremely well to parallelization and that it leads to an effective method for knowledge discovery in complex multidimensional datasets.