NWO Project MacBrain
Machine Learning in Action: Mining the Netherlands Brain Bank

DESCRIPTION

The Netherlands Brain Bank (NBB) contains brain tissue collected from thousands of donors either healthy or aected by neuro-degenerative diseases, including Alzheimer's Disease. It contains also a host of yet un-utilized data describing clinical and pathological information about the donors.
It is a challenging task to exploit these data for discovering computational disease profiles characterized by few features, that could be utilized by domain experts for performing targeted studies in order to improve prognosis, diagnosis and treatment of neuro-degenerative diseases.
Clearly, only clinical features can be used to characterize these profiles, since pathological features are collected after the death of a donor. Nevertheless, pathological data can be used to bias the construction of such disease profiles. Thus we are faced with the scientic problem of what in Machine Learning is called sparse data clustering (here: clustering the clinical data) using privileged information (here: the pathological data). The scientic goal of this project is therefore the development of a flexible and general algorithmic framework for sparse clustering with privileged information. Our approach will consist of three main steps: constraint generation, metric learning, and clustering, allowing for the development and comparative analysis of new algorithms in combination with existing techniques, including our own previous work.
Applying this framework to the NBB data will produce disease profiles described by few clinical features linked to pathological parameters of interest. These profiles will be evaluated and interpreted by the domain experts of our research team, and utilized as a basis for targeted studies on neuro-degenerative diseases.

Contact

 Elena Marchiori
Email: elenam AT cs DOT ru DOT nl

Team