Performance¹ comparison among various methods for querying simulated microarray gene expression dataset. Best results are displayed in bold.

1Performance was measured by the proportions of true positives among the top T genes. T is the number of true positives in each simulated dataset. The mean and standard deviation of these proportions in the 50 simulated datasets were reported. *There are four sub-cases in each of the simulated cases with the same amount of foreground columns. Sub case I: no linear transformation, no cell-level noise; Sub case II: only add linear transformation; Sub case III: only add cell-level noise; Sub case IV: add both linear transformation and cell-level noise. aQuery method using Pearson correlation coefficient. bQuery method using Spearman correlation coefficient. cQuery method using Kendall’s τ. dQuery method using QDB. eQuery method using mutual information. fQuery method using BEST. gQuery method using BEST allowing exclusion of individual cells from the foreground. hQuery method using BEST when fixing the indicator variables of five true target genes and five true experimental conditions as 1.