Fig. 2

Comparison of multi-way algorithms on simulated data. a Sensitivity (SE) versus noise level (x-axis) for seven different methods as indicated, as evaluated on simulated data (data points are averages over 1000 Monte Carlo runs). In each case, the data tensor was of size 2×100×1000, i.e. two data types, 100 samples and 1000 genes. b Left panel: Box plots of SE values for the same seven methods for the largest noise level (5). Each box contains the SE values over the 1000 Monte Carlo runs. Right panel: Corresponding heat map of P values of significance for each pairwise comparison of methods. P values were computed from a one-tailed Wilcoxon rank sum test. For each entry specified by a given row and column, the alternative hypothesis is that the method specified in the row has a higher SE than the method specified in the column. c,d As a,b, but for the specificity (SP). SE sensitivity, SP specificity