Figure 3
Validation framework overview. Module (A) identifies all the genes with at least one variant discovered to be associated with the given disease by the proposed framework. We refer to this list of genes as the proposed variant-driven gene panel. Module (B) first analyzes several independent gene expression datasets studying the given phenotype. We use cross validation method. In each round of sampling, we use one of the gene expression datasets as the training dataset and we use the rest as the testing datasets. We use the expression values of the genes included in the proposed gene panel as the features to build a classifier. Then, we apply the trained classifier on each of the testing datasets in order to predict the patients’ clinical outcome in each testing dataset. We use the area under the curve (AUC) of the receiver-operator characteristic to assess the performance of the classifier. We repeat this procedure n times (where n is the number of gene expression datasets). An average of AUCs is calculated over the n rounds of sampling. This procedure is used to compare the diagnostic quality of the proposed variant-driven gene panel with the current available variant-relevant gene panels obtained from literature.
