Novel Methodology for CRC Biomarkers Detection with Leave-One-Out Bayesian Classification

Abstract

In our previous research we developed a methodology for extracting significant genes that indicate colorectal cancer (CRC). By using those biomarker genes we proposed an intelligent modelling of their gene expression distributions and used them in the Bayes’ theorem in order to achieve highly precise classification of patients in one of the classes carcinogenic, or healthy. The main objective of our new research is to subside the biomarkers set without degrading the sensitivity and specificity of the classifier. We want to eliminate the biomarkers that do not play an important role in the classification process. To achieve this goal, we propose a novel approach for biomarkers detection based on iterative Bayesian classification. The new Leave-one-out method aims to extract the biomarkers essential for the classification process, i.e. if they are left-out, the classification shows remarkably degraded results. Taking into account only the reduced set of biomarkers, we produced an improved version of our Bayesian classifier when classifying new patients. Another advantage of our approach is using the new biomarkers set in the Gene Ontology (GO) analysis in order to get more precise information on the colorectal cancer’s biomarkers’ biological and molecular functions.