In December '09, I organized a 2 hours tutorial session on R SDisc called Hands on R SDisc. A dozen of clinical and computer science researchers attended the session.

In the last years, I served as program committee member for:
IEEE-EMBC'09 and '10
IFIP-AI'08 and '10
ECML-PKDD'09
and I reviewed for the Data Mining and Knowledge Discovery journal.

Recent talks:
- University Medical Center of Utrecht in November '09, pharmacogenomics.

The time and the expertise to perform robust subtyping inferences in data are often regarded as limiting factors for the range of analysis hypothesis considered.

Not only competence in cluster analysis is required but also in exploratory data analysis, regression, statistical testing, computational statistics, classifier training and testing, data visualization and scientific programming. Identifying data subtypes is therefore greatly interdisciplinary.

As graduation project, I work in text mining under the guidance of Prof. P. Brazdil ('05, LIAAD, Porto, Portugal) [1].

For my doctoral dissertation supervised by Prof. J.N. Kok ('06-'09, LIACS, Leiden, the Netherlands) [2], I research data mining scenarios capable of breaking down the complexity of diseases presenting clinical heterogeneity. I mainly collaborated with researchers from the LUMC in the Netherlands.

References

1. Colas, F. 
   2005.  Propositions de Recommandations en Classification de Documents (Giving Advice for Document Classification Tasks).
2. Colas, F. 
   2009.  Data Mining Scenarios for the Discovery of Subtypes and the Comparison of Algorithms.

Interface Magazine, Fall 2009

The identification of homogenous patient subgroups for diseases presenting clinical heterogeneity is of major interest. Osteoarthritis, Parkinson’s disease or major depressive and anxiety disorders represent illustrative examples. Finding subtypes may be helpful in unveiling pathogen mechanisms and subsequently in developing tailored prevention strategies and therapies.

Read the rest of the article in the attached PDF.

Netherlands Bioinformatics Center: http://www.nbic.nl

R SDisc is an integrated set of tools and methods to identify homogeneous subtypes in data distribution by cluster analysis.

The data mining scenario includes methods:
. for data treatment and pre-processing,
. for repeated cluster analysis, model selection, model reliability and reproducibility assessment,
. for profiles characterization and validation by visual and table summaries.

SDisc applies particularly well to subtyping of patient profiles in clinical research.