Research On Tumor Prediction Models Based On Gene Expression Data

AbstractThe analysis and research on gene expression data is an important research area of bioinformatics.It is very significant for detecting and recognizing tumor to establish tumor classification andprediction models by using methodology and technique of information science based on geneexpression data. Artificial Neural Network is applied to create tumor classification and predictionmodels based on gene expression data for Multiple Myeloma, Small Round Blue Cell Tumor andColon. Research results are gained as follows.For Multiple Myeloma, BP Prediction Model(BPPM) is created by using Back Propagation(BP)network. 44 feature genes are selected using correlation analysis and a predictor is designedbased on three layers BP network. Research results indicate that BPPM predicts correctly all oftest samples and correct rate is 100%. Self-Organization Feature Map(SOFM) network is firstlyapplied to research on Multiple Myeloma and Self-Organization Prediction Model(SOPM) withself-organization characteristics is proposed. SOFM network is served as a core of SOPM andself-organization unit is designed to embody self-organization characteristics. Research resultsindicate that prediction correct rate of SOPM is up to 100%.As to Small Round Blue Cell Tumor, the paper takes into account four families of Small RoundBlue Cell Tumor and adopts the thought of dividing multi-class classification into binary classclassification to create Multiple Models Prediction Model(MMPM) based on four BP networks.MMPM is composed of four BP networks, each of which classifies one of four families.Research results indicate that MMPM exactly predicts all of 20 test samples and correct rate ofprediction is up to 100%.For Colon tumor, the paper first applies Learning Vector Quantization(LVQ) algorithm toestablish Colon Learning Vector Quantization Prediction Model(LVQPM). A predictor is createdby using LVQ algorithm and selecting 50 feature genes. LVQPM classifies correctly 57 of 62samples, correct rate of classification is 91.9%. Compared to 90.3% correct rate of S. Furey,88.7% correct rate of Ben-Dor and 87.1% correct rate of Alon, the performance of ColonLVQPM is the best.A part of research results of the paper are published on Chinese Journal of BiomedicalEngineering, Journal of Beijing University of Technology and Proceeding of the 22nd ChineseControl Conference.The paper is supported by National Natural Science Foundation of China.