R2KS: A Novel Measure for Comparing Gene Expression Based on Ranked Gene Lists

Bioinformatics analyses frequently yield results in the form of lists of genes sorted by, for example, sequence similarity to a query sequence or degree of differential expression of a gene upon a change of cellular condition. Comparison of such results may depend strongly on the particular scoring system throughout the entire list, although the crucial information resides in which genes are ranked at the top of the list. Here, we propose to reduce the lists to the mere ranking of the genes and to compare only the ranked lists. To this end, we introduce a measure of similarity between ranked lists. Our measure puts particular emphasis on finding the same items near the top of the list, while the genes further down should not have a strong influence. Our approach can be understood as a special version of a two-dimensional Kolmogorov-Smirnov statistic. We present a dynamic programming algorithm for its computation and study the distribution of the similarity values. The performance on simulated and on real biological data is studied in comparison to other available measures. Supplementary Material is available online (www.liebertonline.com/cmb).