Gene Ontology Driven Feature Filtering from Microarray Data

Microarray data is high-dimensional and noisy. Dimension reduction, i.e. selecting a small number of genes, is an effective way to improve mining efficiency. We propose a novel approach that integrates gene ontology knowledge at the level of feature selection into microarray data to improve binary class prediction. The advantage of this filtering approach lies in considering gene-to-gene relations and selecting more meaningful features comparing to the methods evaluating genes in isolation. In addition, gene ontology knowledge can overcome the limitations of noisy microarray data. Our approach is evaluated on a real benchmark dataset.