A FastMultivariateFeature- Selection/Classification Approach for Predictionof Therapy Response inMultiple Sc erosis

Recombinant Interferon Beta(IFN,) isoneofthemost commonly prescribed treatments formultiple sclerosis; however, thetreatment results inpartial success producing nobenefit in almost halfofthepatients. We address theproblem ofidentifying minimal androbust sets ofmolecular biomarkers thatareableto present predictive modelsofresponse totreatment inmultiple sclerosis patients. To achieve this, we utilize a multivariate featureselection and classification framework; OSeMA (orthogonal searchmodelanalysis) integrates fastorthogonal search algorithm forfeature selection anddiscriminant analysis for classification. Feature-selection and classification performance ofOSeMA areevaluated throughcomparative studies with two wrapper-approachfeature- selection/classification systems.Itisdemonstrated thatthe feature-selection of OSeMA significantly reducesthe computational timeofexhaustive searches whileidentifying complex gene-gene relationships. Utilizing OSeMA,weareable toconstruct classification modelsthatarehighly predictive of therapy response inMS patients, basedontheir geneexpression dataacquired prior toinitiation ofIFN,treatment.