Fig. 1

Automatic analysis framework of oCEM. Gene expression data first underwent the two pre-processing steps: removal of outlier samples and normalization. Then, the user could refer to the recommendation of oCEM regarding which decomposition method should be selected and how many component numbers were optimal by using the function optimizeCOM. Next, the processed data were inputted into the function overlapCEM, rendering co-expressed gene modules (i.e., Signatures with their own kurtosis ≥ 3) and Patterns. Kurtosis statistically describes the “tailedness” of the distribution relative to a normal distribution. Finally, corresponding hub-genes in each module and the association between each module and each clinical feature of choice were identified