A muti-SVMs design for cancer diagnosis using DNA microarray data

Microarray data of gene expression pattern provide useful information for the diagnosis of certain diseases. However the dimension of microarray data is always very high and the volume of samples is small. How to select informative genes remains a challenge. In this research, multiple support vector machine (MSVM) were designed for disease diagnosis. Each SVM was trained using a few gene features. The importance of genes was evaluated by the structure error loss. SVMs with most important genes were linearly combined to form the disease classifier. The algorithm was applied to an artificial dataset. The human acute leukemia dataset was used as a test case.