How to discriminate between potentially novel and considered biomarkers within molecular signature?

The lack of consensus among reported molecular (gene, protein, regulatory marker) signatures (MSs) in the literature is often an initial concern for researchers and subsequently it discourages larger scale prospective studies, prevent the translation of such knowledge into a practical clinical setting and ultimately hindering the progress of the field of biomarker-based disease classification, prognosis and prediction. Understanding the high level of clinical and biological heterogeneity in patients´ cohort distribution, (e.g. by diseases subtypes and stages, age, treatment methods etc), limitations and misbalances in the number of samples, and uncertainty in the dimensionality of potential biomarker space, are critical for getting the signature consensus and identification of novel potential biomarkers. Differences in use of technological platforms, as well as variations in experimental protocols in different studies are also often contributing factors in the lack of strong consensus among signatures. In view of these differences, it would be inappropriate to compare MSs in entirety. Here, we investigate each variable in the signature of interest, and attempt to generate computationally “a null frequency distribution” of the expected number of co-occurrences in other MSs, i.e. other published MSs, and identify both novel and common biomarker within the given MS. We demonstrated an application of proposed model to identification of clinically essential genes of our somatically mutated genes in breast cancer.