Experimental condition selection in whole-genome functional classification

Microarray technologies enable the quantitative simultaneously monitoring of expression levels for thousands of genes under various experimental conditions. This is new technology has provided a new way of learning gene functional classes on a genome-wide. Previously, lots of unsupervised clustering methods and supervised classification have shown power in assigning functional annotations based on gene coexpression. However, due to the noisy and highly dimensional nature of microarray data and the inherent heterogeneity of gene functional classes, the whole-genome learning of gene functional classes from microarray data has remained a great challenge for scientists. Currently, most of the methods do not discriminate the different attribution of experimental conditions in the learning process, which impaired the ability of learning functional classes and prevented these methods from discovering the links between the experimental conditions and gene functional classes. In this study, we perform a selection of experiment conditions during the systematically learning of ∼100 functional classes categorized in MIPS´s comprehensive yeast genome database. In particular, a hybridization of genetic algorithm and k-nearest neighbors classifier has been adopted here. Through a comparison of the results with other previous methods our studies indicate promising improvements in learning performance. Further, by identifying the critical experimental conditions, significant links between the experiments and the functional classes were uncovered.