Network-based biomarkers in Alzheimer's disease: review and future directions

Abstract

By 2050 it is estimated that the number of worldwide Alzheimer's disease (AD) patients will quadruple from the current number of 36 million people. To date, no single test, prior to postmortem examination, can confirm that a person suffers from AD. Therefore, there is a strong need for accurate and sensitive tools for the early diagnoses of AD. The complex etiology and multiple pathogenesis of AD call for a system-level understanding of the currently available biomarkers and the study of new biomarkers via network-based modeling of heterogeneous data types. In this review, we summarize recent research on the study of AD as a connectivity syndrome. We argue that a network-based approach in biomarker discovery will provide key insights to fully understand the network degeneration hypothesis (disease starts in specific network areas and progressively spreads to connected areas of the initial loci-networks) with a potential impact for early diagnosis and disease-modifying treatments. We introduce a new framework for the quantitative study of biomarkers that can help shorten the transition between academic research and clinical diagnosis in AD.

Keywords: Alzheimer’s disease; default-mode network DMN; network degeneration hypothesis; network-based biomarkers; resting-state functional connectivity.

Figure 1

Seven biomarkers of interest are listed in BM. For convenience, we assume that BM is a binary vector, that is, BM(i) = 0,1. For example, if the measurement of the biomarker Word recognition reaches the positive threshold BM(1) = 1, if not, BM(1) = 0. The table in the top of the figure shows the training set S consisting of n samples or subjects with their biomarkers BM, and diagnosed as AD or healthy. The data in the table can be summarized via the construction of generative networks, one for each diagnostic category, in our example H and AD. There is a number of possible network structures that can characterize the training set, so the generative networks M_H and M_AD are the result of model selection. The diagnosis of new patients can be thus be addressed via the computation of the probability that the new data, BM_s is generated by the biomarker network that captures the dependencies among biomarkers in healthy subjects or by the biomarker network of healthy subjects.