Online Microarray Analysis Tool using a modifiedsupport vector machine (MSVM)An internship report for CBS MS DegreeCommittee:Dr. Rosemary Renaut1Professor, Department of Mathematics and Statistics,Director, Computational Biosciences PSMArizona State UniversityDr. Huan Liu2Professor, Department of Computer Science & Engineering,Arizona State UniversityDr. Hongbin Guo3Post-doctoral Fellow, Department of Mathematics and Statistics,Arizona State University Student:Wang-Juh Chen (Sting)4Computational Bioscience PSMArizona State UniversityMay 9, 2005Report Number: 05-031email: [email protected]: [email protected]: [email protected]: [email protected] is becoming an important tool for monitoring and analyzing gene expressionprofiles from thousands of genes simultaneously. Due to the characteristic of these datasets, where thedimension of the feature space is far greater than the sample size, we can not use traditional methodsfor information retrieval. Researchers are looking for some other tools to solve this problem, such assupervised machine learning and data mining. Supervised machine learning and data mining tools arepopular for the analysis of gene expression microarray data. The Support Vector Machine (SVM) isone of these. For the non-separable case, SVM introduces slack variables in mapped space asmeasurements of misclassifications, the source of which may be mislabeling, error of data or/andoutliers. Here we investigate a novel way to approach the misclassifications. We account for themisclassifications in feature space by an errors in data approach, based on the observation of the noisycharacteristic of microarray data. In this internship, we propose a SVM as a classifier, in which themeasurement errors are incorporated in the feature space.During the training phase, involving huge datasets with large number of features but actualsmall sample size, it will cost a lot of time and memory, so we will introduce a CLUSTER platform,ROCKS, to distribute computations. Meanwhile, an online submission interface will be included tomake this tool easier for access.The modified SVM algorithm demonstrates the advantages of choosing Support Vectors, doingthe regularization, and handling the errors in the dataset. These advantages make the algorithm suitablefor different kinds of datasets and tolerant to errors within a dataset, which occurs often duringmicroarray experiments.Keywords : Microarray, SVM, CLUSTER, LAMP2.Goals of internshipDevelop an online tool to analyze microarray data. Implement a modified support vectormachine (MSVM) for classifying microarray data more accurately, and apply CLUSTER system andonline submission for the service.3.Introduction and overview3.1.MicroarrayDNA microarray is a high-throughput tool for rapidly analyzing the expression of all genes withina genome[1,2,3].With thousands or ten of thousands of genes on the microarray chip, we can determine theexpressed genes from a few interesting samples. The most commonly used microarray chip wasdeveloped in the laboratory of Patrick Brown at Stanford University, and then sold commercially by thecompany Affymetrix[1]. The chips are constructed by robotic spotting of DNA fragments, derived fromthe PCR amplification of entire genes, onto precise points of a glass slide (Figure 1). Consequently, theDNA on the chip is fixed by UV cross-linking and then denatured to make the DNA single stranded.The prepared chips are then used as templates for the binding of labeled cDNA fragments. cDNAfragments are derived from the RNA or mRNA of interesting target cells. Those samples could beRNA or mRNA which are virtually correlated with all changes in cell state or type. After binding thefluorescents (for example, sample 1 labeled with Cy3, and sample 2 labeled with Cy5) and then thereverse transcription, RNA samples turn into cDNA, which can hybridize with the single strand DNAon the chip (Figure 2). The color appearing on each spot, expresses the ratio of sample 1 and 2 cDNAbinding with the single strand DNA of the chip. After the samples are accurately assessed and thefluorescent signals are calculated, the expressed genes under different conditions of a genome-widescale can be determined.So far, there are lots of biologists using microarrays to do the transcript profiling in many species.For example, DNA microarray has been used to measure the relative expression of all genes in yeastduring the diauxic shift. Around 50 percent of the differentially expressed genes identified bymicroarray analysis had no previously characterized function[1].Figure 1 Construction of a DNA microarrayResource:Richard J Reece. Analysis of Genes and Genomes. 2004Figure 2 Sample preparation and hybridization with microarrayResource:Richard J Reece. Analysis of Genes and Genomes. 2004Using microarrays is as casting the net in the ocean, you can catch many interesting expressedgenes which have not been found or explored in prior research, and eliminate those genes which are notrelevant to your topic. Microarray technology is also used in cancer research, stem cell research, andDrug Discovery[1,2,3]. In breast cancer and lung cancer research, microarray analysis has allowed sets of genes to beidentified that maybe involved in the process of cellular proliferation during cancer growth, through theexpression differences between normal and cancerous cells. On the other hand, the use of microarraysin cancer treatment is for the classification of cancerous cells based upon the genes they express. Forinstance, the assignment of previously unidentified subtypes to malignant melanoma cells can beinferred through their gene expression pattern. These subtypes can be used to predict the clinicaloutcome of individual melanomas. There are some stem cell researchers using microarrays to transcript the profiles of different typesof stem cells[1]. This has revealed the presence of approximately 200 genes that are expressed withindifferent stem cells that are not expressed within differentiated cells.Microarray applications include drug Discovery and pharmacogenomics. Identifying drug targetsprovided the initial market for the microarray[2]. A good drug target has extraordinary value fordeveloping pharmaceuticals. By comparing the differentiations in which genes are expressed in normaland diseased tissue, scientists might be able to identify the genes and hence the associated proteins, orthe proteins with which drugs