Figure 3.

Comparison with previous studies. Each value corresponds to a sum of log-odd scores from up and down regulated gene sets. In prognosis analyses, pattern based methods (MEGA, multiple regression and ANOVA models) showed better results than two-class comparison methods (Paired t-test in Suzuki and Mann-Whitney test in Ellsworth). N-MEG (blue) showed the best significance among all the gene sets. Instead, N-MEG and a multiple regression set showed relatively low significance in tumor stage data; probably because most of the N stage test sets used two-class comparison methods. Values in tumor grade analyses were scaled down to 1/10 for better presentation of the graph. N-MEG = N-wise monotonically expressed genes, MR = multiple regression, 2ANOVA = two-way ANOVA.