The search for biomarkers for response to a therapy consists of several steps. It starts with the acquisition of data and biological samples from a number of patients prior to the therapy. The collection of these data and samples form a biobank. After an appropriate time following the therapy, the response data are integrated in the biobank and employed to distinguish responders from non-responders. In the next step, the samples from all patients are analyzed, resulting in the identification of some features specific to responders and other features specific to non-responders. All these features are considered to be biomarker candidates. In a further step, the biomarker candidates are evaluated in validation studies involving new patients; if they succeed in predicting therapy response in these studies, they can be employed in the future as biomarkers for response. A similar workflow is used in the search of biomarkers for prognosis and for adverse reactions.
In recent years, many biomarker candidates for cancer prognosis, therapy response, and adverse reactions of a therapy have been identified. Unfortunately, the majority of them have failed in validation studies due to their poor reproducibility and/or their poor accuracy in the medical prediction1,2. There are good reasons for this: