Performance comparison of gene-expression-based classifiers for GBM patient prognosis.

<p><b>A</b>) Prognostic accuracy of various marker sets. Classification accuracy is defined as the ratio of the number of correctly classified patients to the total number of patients tested. Expression data of 42 GBM patients was used to derive the eModule set. Top-gene set is top 156 (size-matched to the number of genes in the eModule set) most significantly differentially expressed genes between LTS and STS patients. 38-gene set, a set of 38 discriminative genes reported in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0052973#pone.0052973-Colman1" target="_blank">[4]</a>. Two hundred thirty seven additional GBM patients from TCGA were used for testing classification accuracy. Error bar is the standard deviation based on 100 leave-one-out cross validations. <b>B</b>) Performance of eModule set and 38-gene set using three external microarray data <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0052973#pone.0052973-Freije1" target="_blank">[28]</a>–<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0052973#pone.0052973-Phillips1" target="_blank">[30]</a> from which the 38-gene signature was derived. Numbers in parenthesis indicate number of LTS and STS in each data set, respectively. P-values are based on t-tests comparing the average classification accuracy of the eModule-based classifier and those of other classifiers.