Identification of significant biological markers at metabolic and genomic level for non - invasive discrimination of brain tumors

Abstract

Cancer is generally considered as the public nuisance of our century since it is one of the most complex diseases that the medical community must face. Its undetermined pathological origins, its unpredicted biological behavior and its lethal, most of the times, outcome are some of its main characteristics that experts have to deal with. Among the most lethal types of cancer is the Brain cancer which is characterized by the formation of one or more solid tumors within the brain parenchyma. The Brain tumors have the ability to rapidly progress form low malignancy to high malignancy, restricting thus the oncologist’s ability to accurately evaluate their behavior and design an effective treatment in order to improve the patient’s clinical image. The recent release of the human genome enabled experts to understand that abnormal genetic mutations are the basis for cancer genesis. Furthermore, the introduction of other “omics” fields such as transcriptomics, proteomics and metabolomics, descendants of genomics, revolutionized the way experts analyze Brain tumors today. State of the art “omics” technologies and pattern recognition methods managed to revealed useful information regarding Brain tumors’ pathology that has been unknown for many decades. Although a lot has already been done in the field of Brain cancer diagnosis, prognosis and treatment, a lot more must be achieved. Most of the patients with high grade brain malignancy die within 24 months from initial diagnosis. The need to design new and more effective treatments that will prolong patients’ life expectancy is overwhelming. Motivated by this need under the major hypothesis that the selection and the effectiveness of the therapy to be followed is primarily based on the estimation of the histopathological profile of the tumor at diagnosis stage, we attempt to identify novel and reliable biological features (markers) sets that can be adopted to accurately discriminate the type and grade of a brain tumor for a new patient. The selection of significant features, which describe the tumors’ type and grade, is the foundation for the design of novel non-invasive patient specific therapies. This is actually an open challenge in this continuous fight against Brain as well as other types of cancer. To accomplish this goal, the data of Brain cancer patients provided from two “omics” technologies, named Magnetic Resonance Spectroscopy (MRS) and DNA Microarrays, were utilized. MRS technology reveals the metabolic profile of a Brain tumor while DNA Microarrays provides its genetic identity. Analyzing the information provided from MRS spectra, we identified novel metabolic marker sets that can be used to classify the type and grade of a new patient with high accuracy. On the other hand our genetic analysis was based on the Otto Warburg’s hypothesis in 1956 who observed that tumorous cells exhibit increased rates of glycolysis (sugar splitting process for cellular energy production). Examining the glycolytic profile of Brain tumor patients we managed to discover that, apart from the well known from bibliography genetic markers (genes), glycolycis related genetic markers play a very significant role in Brain tumor’s behavior. Based on this two-fold analysis a novel medical Decision Support System (DSS), which bridges the knowledge extracted from two different “omics” modalities, i.e. genomics and metabolomics, is proposed. As a primary result, we verify the importance of metabolites in cancer-type and grade discrimination and validated their metabolic and genetic association in cancer progression, through the glycolysis process. In order to implement the analysis of the data at genomic and metabolomic levels, modern pattern recognition methods were applied. Two well known classifiers named Support Vector Machines (SVM) and the Least Squares-SVM (LS-SVM), widely used in biomedical problems, were used exploiting their unique property to cope quite well with complex data as occurs in brain cancer. Based on these classifiers we managed to develop a reliable feature selection and classification system that embeds the intrinsic characteristic of patients’ data into the classification process resulting to high classification accuracy rates and identification of significant metabolic and genetic marker sets. This was a secondary accomplishment of this thesis