| Abstract: |  Classification of mRNA expression data in order to determine the presence of a severe disease is an active topic of bioinformatic research. In recent years, the newly-discovered microRNAs, a group of compounds inhibiting mRNAs, have been gaining prominence in these tasks.
The thesis explores the current trends in integration of mRNA and microRNA data and the effect of integration on the quality of class prediction. The thesis also proposes two new integration methods which use microRNA-mRNA target prediction databases as prior knowledge — one based on dimensionality reduction using singular value decomposition, one based on subtracting a proportion of microRNA expression levels from the expression levels of the related inhibited mRNAs.
The existing and newly proposed integration methods are experimentally evaluated on myelodysplastic syndrome data. The experiments show that the models trained on datasets integrated using the new methods outperform both the models trained on datasets integrated trivially and the models trained on non-integrated datasets.
|
|---|