Regression-Based Cancer Metabolism Analysis And Gene-Gene Relatedness Study

Regression is a kind of algorithms and models for estimating the relationships among variables.Now its applications have been leading to a great-leap-forward development in various fields,especially in bioinformatics.In this study,we creatively employ,design and develop regression-based models to analyze biological Omics data and address bioinformatics problems.Our study mainly focuses on two topics,including glutamine and glutamate metabolisms analysis in different cancers and regression-based gene-gene relatedness study.In the first part,unlike cell-line and animal model-based studies,we conduct a comparative analysis of gene-expression data of cancer vs.normal control tissues of 11 cancer types,which can exactly reflect the characteristics of glutamine and glutamate metabolisms in human cancer trusses;and we creatively adopt a multiple multivariate linear regression model to assess differential contributions by glutamine and/or glutamate to each of seven biological processes in cancer vs.control tissues,which represents the first comprehensive study of glutamine and glutamate involvements in the selected metabolic pathways in cancer vs.control tissues.Through systematic analysis,we acquire a novel understanding of glutamine and glutamate metabolisms in cancer.Specifically,we comprehensively predicted the enhancement roles of glutamine and glutamate metabolisms in the selected metabolic pathways in cancer vs.normal tissues,which may promote the targeted therapeutic development of glutamine and/or glutamate metabolism.In the second part,we design and develop a novel regression-based multi-feature relatedness(MFR)model to measure gene-gene relatedness under specific experimental conditions by integrating co-expression similarities and prior-knowledge similarities,and by balancing both co-expression relatedness and prior-knowledge relatedness.We propose three important hypotheses and use support vector machine as a core part of MFR model to realize the multi-objects optimization and make MFR model offer excellent capability of generalization.In order to solve the problem that using support vector machine,a generalized linear logit regression model,as a core part of our novel model for measuring the continuous attribute gene-gene relatedness,we utilize the “probability” of a gene-pair sample belonging to the positives or negatives,namely MFR,which evaluates gene-gene relatedness by computing the normalized distance from the gene-pair sample to the optimal hyperplane of the support vector machine.Comparing with other multi-feature linear models and co-expression analysis methods,MFR model has demonstrated the best precision,robustness and validity.The script of MFR model is implemented by R language and can be downloaded freely at MFR model web site for academic uses.The two parts of our study not only are based on regression theories but also represent the two important applications of regression in specific bioinformatics problems resolving,will be combined in the further research,and their results are of great practical value.