Network-based integration of clinical, imaging and molecular biomarkers of dementia

Biomarkers are measurable characteristics used as indicators for diagnosis, prognosis and therapy-response: they can be of different nature (e.g., clinical, imaging and molecular) and each type captures and describes one or few specific aspects of disease development and progression. However, a single biomarker is rarely enough to operate the high-level characterization needed in a precision medicine scenario. Still, it is not a common practice to study different biomarkers of the same disease at the same time, focusing on their inter- relationships and role in the disease. The common practice is to evaluate pairwise relations. Thus, the definition of comprehensive and integrative biomarkers is an open challenge especially in complex diseases, such as in dementia. Dementia is in fact a multi- factorial condition, with factors related to different domains and whose interactions are not well known. In this study, we applied simple and partial correlation to data of patients and healthy subjects related to a set of relevant features of dementia. Network representation and analysis showed that both approaches provide a clear pattern of interactions among different heterogeneous biomarkers. We found that, even if the two tested methods agree on the strongest edges, partial correlation graphical LASSO can detect more connections and is more sensible to negative correlations providing a more complete description of biomarkers interactions. Comparing the biomarker networks of patients and control cases we also highlighted the changes that can be seen with the development of the disease.