Comparative study of two-layer particle swarm optimization and particle swarm optimization in classification for tumor gene expression data with different dimensionalities

Classification of gene expression data to determine different type or subtype of tumor samples is significantly important to research tumors in molecular biology level. Sample genes (dimensionalities) play a fundamental role in classification. Feature selection technologies used to reduce gene numbers and find informative genes have been presented in recent years. But the performance of feature selection in gene classification research is still controversial. In this study, a classification algorithm based on the two-layer particle swarm optimization (TLPSO) is established to classify the uncertain training sample sets obtained from three gene expression datasets which contain the leukemia, diffuse large B cell lymphoma (DLBCL) and multi-class tumors dataset respectively with the exponential increasing of gene numbers. Compared the results obtained by using the particle swarm optimization (PSO), the classification stability and accuracy of the results based on the proposed TLPSO classification algorithm is improved significantly and more information to clinicians for choosing more appropriate treatment can extracted.