Fig. 1

ADAGE signature analysis workflow. An input dataset is first processed to be compatible with a provided ADAGE model trained on the provided expression compendium. Raw microarray measurements are transformed by RMA, while processed data first have missing values imputed and then are normalized to the compendium. The second step is detecting active signatures in the dataset. Signature activity for each sample in the dataset is computed. Statistical test on the signature activity is used to identify signatures respond to an experimental treatment. Redundant signatures are further removed. The last step is interpreting the biological meaning of active signatures through evaluating genes in the signatures, pathways associated with the signatures, datasets related to the signatures, and a gene-gene network built upon the signatures