Fig. 1

General idea of CL. Due to condition-specific biases, the existing normalization algorithm might fail to normalize the distributions of class-specific gene signatures in the reference and prediction datasets. Therefore, the classifier trained using the reference dataset would not work well for the prediction dataset (top right figure). Unlike normalization based approach, CL exploits the fact that the signature is unique to its associated class under any condition and thus employs an unsupervised clustering algorithm to discover this unique signature, hence the class label (bottom right figure)