Search

Gene microarray experiment can monitor the expression of thousands of genes simultaneously. Using the promising technology, accurate classification of tumor subtypes becomes possible, allowing for specific treatment that maximizes efficacy and minimizes toxicity. Meanwhile, optimal genes selected from microarray data will contribute to diagnostic and prognostic of tumors in low cost. In this paper, we propose an improved FMM (fuzzy Min-Max) neural network classifier which provides higher performance than the original one. The improved one can automatically reduce redundant hyperboxes thus it can solve difficulty of setting the parameter θ value and is able to select discriminating genes. Finally we apply our improved classifier on the small, round blue-cell tumors dataset and get good results. [ABSTRACT FROM AUTHOR]

As the gene sequencing technology has been maturating, genomes of more and more organisms and genetic sequences are available in international repositories. However, the biological functions of most of these genes remain unknown. Microarray technology opens a door to discover information of underlying genes and is widely used in basic research and target discovery, biomarker determination, pharmacology, target selectivity, development of prognostic tests and disease-subclass determination. Clustering is one of the typical methods of analyzing microarray data. By clustering the gene microarray expression data into categories with similar profiles, genes with similar function can be focused on. There are many clustering methods used for the analysis of gene microarray data. However, they usually suffer from some shortcomings, such as sensitive to initial input, inappropriate grouping, difficult to discover natural or near optimal clusters, and so on. In this paper, we propose a novel clustering method to discover the optimal clusters by searching PPVs (Pair of Prototype Vector). The experiment results show that our method works very well. [ABSTRACT FROM AUTHOR]

The construction for a P2P-based high performance computing platform (P2HP) is presented to address parallel problems in this paper. P2HP utilizes idle computers in the Internet with great scalability to form an enormous computing capability for scientific supercomputing and volunteers form autonomous unstructured P2P network domains. The configuration of P2HP is easy and a programming model is provided. Its applications involve a large range of problems, and a benchmark is applied to evaluate its performance. [ABSTRACT FROM AUTHOR]

A stereo vision [1,2] is used to measure the crack tip parameters (such as J integral[3], plastic mixity [4], and elastic mixity [4]) of mixed mode fracture specimens, and to study the applicability of the Shih's plane strain solution [4,5] to the mixed mode crack tip fields. The fracture specimen used in the paper is a compact tension shear (CTS) specimen made of 2024-O aluminum. To conduct mixed mode fracture experiments at the loading angles of 75° and 45°, a special loading device is used [6,7]. Fig. 1 shows the experimental setup for the loading angle of 75°. [ABSTRACT FROM AUTHOR]