Applied Research
In the domain of
biology I have worked extensively on the construction of diagnostic and
prognostic models for biological problems with a focus on proteomics problems
and mass
spectrometry, within the context of a European Cost Action, and dealt
extensively with quality assurance and control issues. The main part of the
work focused on preprocessing and feature extraction followed by the
application of machine learning to model how the extracted features determine
the final patient outcome. Preprocessing issues I dealt with included baseline
removal, denoising, smoothing, peak extraction and alignment, I have developed
methods to select among different preprocessing methods and fine tune their
parameters in order to retain the highest possible information content in the
extracted features. Moreover I have participated in the definition of protocols
to control the reproducibility of different sample preparation methods and
procedures to control the within laboratory reproducibility of measurements and
data analysis results.
On a parallel
direction I have worked on the development of learning methods that are adapted
to the idiosyncrasies of the data typically found in mass spectrometry, namely
the high dimensionality and redundancy, using kernel tools. The resulting tools
have been applied successfully to a variety of problems that share similar
characteristics, such as classification of microarrays or text mining problems,
results have been published in the SIAM conference on Data Mining. In a similar
context, within the European project DropTop,
I am working on multisource learning from proteomics, genomics, and
transcriptomics data for the construction of prognostic models using survival
analysis and exploiting learning methods from the area of kernel combination,
weighting and learning.
In the domain of
health informatics and within the context of the European project, DebugIT, whose major goal is to extract new
knowledge concerning problematic patient-safety patterns which will be
incorporated in the hospitalŐs monitor and decision support tool, I am
investigating the development of data mining methods that are appropriate for
the multimodal and spatiotemporal nature of the sequence of events that take
place during the stay of a patient within the hospital. A number of challenges
have to be addressed foremost among them is the semantic integration of data,
information, and knowledge from a variety of different sources, e.g.
information systems and data repositories of different hospitals. The results
will be integrated into the Clinical Information Systems of participating
European hospitals, industry (AGFA Health Care), and their clients and will
become available globally.