Prediction of Candidate Primary Immunodeficiency Disease Genes Using a Support Vector Machine Learning Approach

Shivakumar Keerthikumar¹^,2^,3, Sahely Bhadra⁴, Kumaran Kandasamy¹^,2^,5, Rajesh Raju¹^,2^,3, Y.L. Ramachandra², Chiranjib Bhattacharyya⁴, Kohsuke Imai⁸, Osamu Ohara⁶^,7, Sujatha Mohan¹^,3 and Akhilesh Pandey¹^,5^,*

¹ Institute of Bioinformatics, International Technology Park, Bangalore 560 066, India
² Department of Biotechnology and Bioinformatics, Kuvempu University, Jnanasahyadri, Shimoga 577 451, India
³ Research Unit for Immunoinformatics, Research Center for Allergy and Immunology, RIKEN Yokohama Institute, Kanagawa 230-0045, Japan
⁴ Department of Computer Science and Automation, Indian Institute of Science, Bangalore 560 012, India
⁵ McKusick-Nathans Institute of Genetic Medicine, Johns Hopkins University School of Medicine, 733 N. Broadway, BRB Room 527, Baltimore, MD 21205, USA
⁶ Laboratory for Immunogenomics, Research Center for Allergy and Immunology, RIKEN, Yokohama Institute, Kanagawa 230-0045, Japan
⁷ Department of Human Genome Technology, Kazusa DNA Research Institute, 2-6-7 Kazusa-Kamatari, Kisarazu, Chiba 292-0818, Japan
⁸ Department of Medical Informatics, National Defense Medical College, Saitama 359-8513, Japan

Received 23 July 2009 ; accepted 5 September 2009.

Screening and early identification of primary immunodeficiencydisease (PID) genes is a major challenge for physicians. Manyresources have catalogued molecular alterations in known PIDgenes along with their associated clinical and immunologicalphenotypes. However, these resources do not assist in identifyingcandidate PID genes. We have recently developed a platform designatedResource of Asian PDIs, which hosts information pertaining tomolecular alterations, protein–protein interaction networks,mouse studies and microarray gene expression profiling of allknown PID genes. Using this resource as a discovery tool, wedescribe the development of an algorithm for prediction of candidatePID genes. Using a support vector machine learning approach,we have predicted 1442 candidate PID genes using 69 binary featuresof 148 known PID genes and 3162 non-PID genes as a trainingdata set. The power of this approach is illustrated by the factthat six of the predicted genes have recently been experimentallyconfirmed to be PID genes. The remaining genes in this predicteddata set represent attractive candidates for testing in patientswhere the etiology cannot be ascribed to any of the known PIDgenes.

Key words: RAPID; SVM; HPRD; Human Proteinpedia; NetPath

^* To whom correspondence should be addressed. Tel. +1 410-502-6662. Fax. +1 410-502-7544. E-mail: pandey{at}jhmi.e.du

Edited by Minoru Ko

Add to CiteULike CiteULike Add to Connotea Connotea Add to Del.icio.us Del.icio.us What's this?