Figure 5

Analysis of reproducibility of the detections of the topological algorithms. The reproducibility of the DC analysis of ER- and ER+ breast cancer subtypes is analyzed in two-fold cross-validation. (A) Reproducibility of the score p (DCglob) calculated in a training set (n=140) and in an independent validation set (n=140). Lines mark the threshold (p=0.05) for the detection of differential correlation. (B) Reproducibility of the score d (DCloc) calculated in a training set (n=140) and in an independent validation set (n=140). Lines mark the threshold (d=0.3) for the detection of differential correlation. (C) Reproducibility analysis of the lists of differentially correlated genes. A series of training and independent validation sets of equal size is randomly drawn from the 1035 ER+ and the 282 ER- patients. A gene detected among the top 5% differentially correlated genes in the training set is considered as validated, if it is among the top 5% DC genes in the validation set. (D) As C, but a gene is considered as validated, if it is among the top 20% differentially correlated genes in the validation set.