Figure 1

The key steps of the FLAME algorithm shown on a small simulated dataset ("Starting data"). Step One: expression data are used to calculate for each gene a density value corresponding to the average similarity to its nearest neighbors (in the picture, darkness of each spot is proportional to density); Cluster Supporting Objects (CSOs) are then identified as genes with local maximum density and assigned unique membership to themselves. The red and green colors define two CSOs, while the blue color indicates outliers. Step Two: for all the other genes, a fuzzy membership vector is approximated from the memberships of their nearest neighbors, until convergence; for each spot, red, green and blue colors are now mixed in accordance with the fuzzy membership of that gene to the two clusters or to the outlier group. Step Three: at the end of this process, genes can be assigned to one of the two clusters built around the CSOs or to the outlier group, based on their approximated memberships.