Development of novel multiple comparison method and decision tree method using resampling techniques and its applications to medical and pharmaceutical data.

上个财政年度，我们开发了新的更适用的统计多重比较算法，我们的方法的检验能力被证明是等于或优于传统的不太适用的方法，如Dunnett检验和Tukey检验。我们的研究结果发表在Journal of Computer Aided Chemistry（vol.4，pp.27-34（2003））上。在这一年中，我们尝试应用一种重新排序方法来获得决策树中使用的可预测变量的重要性，这种方法获得了成功。本年度的主要成果如下：1）决策树方法采用了类似于bootstrap方法的重新排序技术，以发现显著的可预测变量（属性）：在决策树方法中，发现属性的显著性是不可能的困难。我们的技术不仅能够找到这样的意义，而且还可以使用一些3D图形可视化的显着性水平。这满足 关于我们 将hod应用于脑肿瘤患者的预后预测，取得了满意的结果。我们的部分结果在由本研究项目的首席研究员（TAKAGI Tatsuya）和Kozo TAKAYAMA组织的日本药学会（大坂）第124届年会的专题讨论会上进行了报告，并在2004年亚洲化学计量学和生物信息学会议（上海）上由本研究项目的首席研究员（TAKAGI）作为特邀演讲人进行了报告。2）基于上一财年的研究结果，我们尝试用递归方法编写快速并行的多重比较法软件，并通过CGI在WWW上发布。该软件已在WWW页面上发布，http：//www.gen-info.osaka-u.ac.jp/testdocs/tomocom/，该软件的手册，http：//www.gen-info.osaka-u.ac.jp/testdocs/tomocom/tazyu.html，以及我们算法的简单解释，http：//www.phs.osaka-u.ac.jp。这个软件经常被使用。少

Last fiscal year, we developed novel more applicable algorithm of statistical multiple comparison, and the power of test of our method was revealed to be equal to or better than the traditional less applicable methods, such as Dunnett's test and Tukey's test. The results of our study were published in the Journal of Computer Aided Chemistry(vol.4, pp.27-34 (2003)). In this year, we tried to apply a resampling method for obtaining the significance of predictable variables used in decision trees, and this approach was succeeded. The main achievements in this year are as described below.1) A resampling technique, which is similar to bootstrap method, was adopted for the decision tree method in order to find significant predictable variables (attributes) : it was impossibly difficult to find the significance of attributes in the case of decision tree methods. Our technique is not only able to find such significance but also visualize the level of significance using some 3D graphs. This met … More hod was applied to the prediction of the prognosis of brain tumor patients and satisfied results were obtained. A part of our results was reported at the symposium at the Symposia of the 124th Annual Meeting of the Pharmaceutical Society of Japan (Osaka) organized by the head investigator (Tatsuya TAKAGI) of this research project and Kozo TAKAYAMA, and at the Conference of Chemometrics and Bioinformatics in Asia 2004 (Shanghai) by the head investigator (TAKAGI) of this research project as an invited speaker. After the conference, this report was published in the Journal of Computer Aided Chemistry (vol.5, pp.35-46 (2004)).2) Based on the results of the studies in the last fiscal year, we tried to write fast and parallel software of our novel multiple comparison method using resampling method and publish on WWW pages by CGI. This software was already published on the WWW page, http : //www.gen-info.osaka-u.ac.jp/testdocs/tomocom/, with the manual of this software, http : //www.gen-info.osaka-u.ac.jp/testdocs/tomocom/tazyu.html, and the simple explanation of our algorithm, http : //www.phs.osaka-u.ac.jp /homepage/b015/research1.html. This software has been being used frequently. Less