A linear programming model for identifying non-redundant biomarkers based on gene expression profiles

With the development of high-throughput technologies, e.g. microarrays and the second generation sequencing technologies, gene expression profiles have been applied widely to characterize the functional states of various samples at different conditions. This is especially important for clinical biomarker identification that is vital to the understanding of the pathogenesis of a certain disease and the subsequent therapies. Because of the complexity of multi-gene disorders, a single biomarker or a set of separate biomarkers often fails to discriminate the samples correctly. Moreover, biomarker identification and class assignment of diseases are intrinsically linked while the current solutions to these two tasks are generally separated. Motivated by these issues, we give out a novel model based on linear programming in this study to simultaneously identify the most meaningful biomarkers and classify accurately the disease types for patients. Results on a few real data sets suggest the effectiveness and advantages of our method.