Genome Wide Functional Annotation Using Mathematical Programming

摘要

Recent advances in cDNA microarray technology have resulted in a surge of gene expression data. Whole genome hybridization results have been reported for many organisms (yeast, worm, etc.) under different experimental conditions of interest. Studies have been reported on various methods for assigning functionality to previously unknown genes. Some of these methods are clustering techniques, self-organizing maps and knowledge based support vector machines (SVMs). These techniques use a similarity measure to associate genes with unknown functionality with genes of known functionality. In this paper we report genome wide functional annotation, for yeast data set, based on an SVM strategy. In our SVM model each gene can potentially be assigned to more than one class. A gene is referred to as a positive sample of a given class if it belongs to that class, otherwise the gene is referred to as a negative sample. We establish the viability of the SVM model using yeast data with several missing experimental measurements. The SVM model correctly annotated the positive samples in all classes. On the other hand the negative genes were not so well annotated. This is attributed to the missing data for the negative gene set.