Work package leader: Brice Ozenne, NRU
In order to meet the clinical need of robust diagnostic and prognostic classifiers for the individual at risk or with a manifest brain disorder, we will in this work package make use of existing data either from our Cimbi database or from collaborators to be analyzed in the context of a multivariate data analysis framework and subsequently test the identified key variables for their predictive value in new data sets. We will use machine-learning techniques and seek to define a set of parallel biomarkers that can optimize the prediction of treatment response with high validity. In addition, statistical assistance for the other three work packages will be provided, both in terms of study design and optimal statistical analyses.
We hypothesize that generation of predictive statistical models will allow for a more informed use of data and will provide a framework for optimized study designs in the future. For this purpose, recent advances in high-dimensional statistics and machine-learning are employed. Depression and certain other brain disorders are characterized by differences in functional brain connectivity as determined by rs-fMRI; this approach may offer a sensitive measure for disease classification. Data acquired in NP1-NP3 will be used to extract resting-state brain networks and we will use multivariate statistical analysis applied to discover networks that, e.g., predict recovery from depression before initiation of drug intervention. A validated prediction model may serve as an important step in translating the knowledge gathered from NP1-NP3 into directly clinically applicable tools.