Article
Analysis of IDDM Rats concerning Diabetes
Search Medline for
Authors
Published: | September 20, 2011 |
---|
Outline
Text
Introduction: The investigation of the mechanism of tolerance induction by the modulatory anti CD4 monoclonal antibody RIB 5/2 in insulin dependent diabetes mellitus rat is the aim of our study. By this approach it should be possible to identify biomarkers of autoimmunity and tolerance for prediction of diabetes onset. So far, we have mainly applied decision trees and some other classification methods on data sets of just twelve specifically breaded rats. Since actually another dozen rats is breaded, we hope to present more sophisticated results soon.
Material and Methods: We used data of the measurement time points: 45, 50, 55, and 60 days of life, because it is biologically not possible to take blood samples every day and this time period is supposed to be most decisive. The attributes are eighteen preselected genes and biomarkers. The class labels are “diabetes”, “no diabetes”, and “background strain”. For classification we used the WEKA tool and applied these prominent classification algorithms: Random Forest, Nearest Neighbour, Naïve Bayes, and Support Vector Machines. However, since we were not so much interested in the classification performance but in those attributes that are best to distinguish between diabetes and no diabetes, we applied decision trees (originally developed as C4.5 [1], later implemented as J48 in WEKA).
Results: The analysis of gene expression patterns might help to distinguish between T1DM affected subjects and healthy animals at an early stage. By applying decision trees, we could demonstrate that analysis of selected genes of T cell differentiation, T cell function, and cytokine expression in whole blood cells at an early prediabetic stage (after 45 days of live), the RT6 T cell proliferation gene was most decisive for diabetes onset in the IDDM rat followed by selectin and neuropilin at the stage of islet infiltration (after 50 days), and IL-4 during progression of beta cell destruction (after 55 days).
Discussion: So far our data set is very small and, probably because of poor data quality, the cross-validated classification results are not so well. Nevertheless, the results of the generated decision trees can be very well explained by the biochemical experts.
References
- 1.
- Quinlan JR. C4.5 Programs for Machine Learning. San Mateo: Morgan Kaufmann; 1993.