Gene selection using support vector machines with non-convex penalty

• DOI: 10.1093/bioinformatics/bti736
• Publication Date: 2005-10-27T00:12:37Z
• Authors: Hao Helen Zhang; Jeongyoun Ahn; Xiaodong Lin; Cheolwoo Park
• Keywords: Models, Statistical; Time Factors; Models, Genetic; Bayes Theorem; Sequence Analysis, DNA; 01 natural sciences; Pattern Recognition, Automated; 3. Good health; Neoplasms; Databases, Genetic; Cluster Analysis; Humans; 0101 mathematics; Algorithms; Software; Oligonucleotide Array Sequence Analysis

With the development of DNA microarray technology, scientists can now measure expression levels thousands genes simultaneously in one single experiment. One current difficulty interpreting data comes from their innate nature 'high-dimensional low sample size'. Therefore, robust and accurate gene selection methods are required to identify differentially expressed group across different samples, e.g. between cancerous normal cells. Successful will help classify cancer types, lead a better understanding genetic signatures cancers improve treatment strategies. Although classification two closely related problems, most existing approaches handle them separately by selecting prior classification. We provide unified procedure for simultaneous classification, achieving high accuracy both aspects.In this paper we develop novel type regularization support vector machines (SVMs) important A special nonconvex penalty, called smoothly clipped absolute deviation is imposed on hinge loss function SVM. By systematically thresholding small estimates zeros, new eliminates redundant automatically yields compact classifier. successive quadratic algorithm proposed convert non-differentiable non-convex optimization problem into easily solved linear equation systems. The method applied real datasets has produced very promising results.MATLAB codes available upon request authors.