| The development of DNA microarray technology has enabled gene expression analysis based on simultaneous observation of thousands of genes. One of objectives of microarray analysis is to discover interesting structures from large amounts of expression data, and hierarchical clustering is often used to accomplish this. However, the estimation is often susceptible by statistical sampling error, and thus the result may be obtained only by chance without reflecting the true hypothesis. Therefore, it is necessary to evaluate the reliability of the hypothesis obtained from the analysis. We applied multiscale bootstrap resampling (Shimodaira 2002 [4], 2004 [6]) to evaluate the accuracy of hypotheses as p-values. This method is based on resampling of data, and applicable to a large class of problems including hierarchical clustering. As an example, we show an application to hierarchical clustering of lung tumor microarray data. |