An Evolutionary Variable Neighborhood Search for Selecting Combinational Gene Signatures in Predicting Chemo-Response of Osteosarcoma

Abstract

In genomic studies of cancers, identification of genetic biomarkers from analyzing microarray chip that interrogate thousands of genes is important for diagnosis and therapeutics. However, the commonly used statistical significance analysis can only provide information of each single gene, thus neglecting the intrinsic interactions among genes. Therefore, methods aiming at combinational gene signatures are highly valuable. Supervised classification is an effective way to assess the function of a gene combination in differentiating various groups of samples. In this paper, an evolutionary variable neighborhood search (EVNS) that integrated the approaches of evolutionary algorithm and variable neighborhood search (VNS) is introduced.It consists of a population of solutions that evolution is performed by a variable neighborhood search operator, instead of the more usual reproduction operators, crossover and mutation used in evolutionary algorithms. It is an efficient search algorithm especially suitable for tremendous solution space. The proposed EVNS can simultaneously optimize the feature subset and the classifier through a common solution coding mechanism. This method was applied in searching the combinational gene signatures for predicting histologic response of chemotherapy on osteosarcoma patients, which is the most common malignant bone tumor in children. Cross-validation results show that EVNS outperforms the other existing approaches in classifying initial biopsy samples.