Machine learning for bioinformatics and neuroimaging

This study describes comprehensive polling of transcription start and termination sites and analysis of previously unidentified full-length complementary DNAs derived from the mouse genome. We identify the 5' and 3' boundaries of 181,047 transcripts with extensive variation in transcripts arising from alternative promoter usage, splicing, and polyadenylation. There are 16,247 new mouse protein-coding transcripts, including 5154 encoding previously unidentified proteins. Genomic mapping of the transcriptome reveals transcriptional forests, with overlapping transcription on both strands, separated by deserts in which few transcripts are observed. The data provide a comprehensive platform for the comparative analysis of mammalian transcriptional regulation in differentiation and development.

To elucidate gene function on a global scale, we identified pairs of genes that are coexpressed over 3182 DNA microarrays from humans, flies, worms, and yeast. We found 22,163 such coexpression relationships, each of which has been conserved across evolution. This conservation implies that the coexpression of these gene pairs confers a selective advantage and therefore that these genes are functionally related. Many of these relationships provide strong evidence for the involvement of new genes in core biological functions such as the cell cycle, secretion, and protein expression. We experimentally confirmed the predictions implied by some of these links and identified cell proliferation functions for several genes. By assembling these links into a gene-coexpression network, we found several components that were animal-specific as well as interrelationships between newly evolved and ancient modules.

Support vector machines (SVMs) were originally designed for binary classification. How to effectively extend it for multiclass classification is still an ongoing research issue. Several methods have been proposed where typically we construct a multiclass classifier by combining several binary classifiers. Some authors also proposed methods that consider all classes at once. As it is computationally more expensive to solve multiclass problems, comparisons of these methods using large-scale problems have not been seriously conducted. Especially for methods solving multiclass SVM in one step, a much larger optimization problem is required so up to now experiments are limited to small data sets. In this paper we give decomposition implementations for two such "all-together" methods. We then compare their performance with three methods based on binary classifications: "one-against-all," "one-against-one," and directed acyclic graph SVM (DAGSVM). Our experiments indicate that the "one-against-one" and DAG methods are more suitable for practical use than the other methods. Results also show that for large problems methods by considering all data at once in general need fewer support vectors.