ゲノム・プロテオミクスデータを用いた予測解析：機械学習による新しい統計的手法

At the present day, it becomes imperative to develop appropriate statistical methods for high-dimensional and small sample data analysis because data formats in the biological or medical fields have been dramatically changed. Especially, it will be common in the near future to analyze clinical data together with genomic data. In this review paper, we introduce several current approaches to the analysis relating to genomic and proteomic data, and describe some limitations or problems in the statistical performance.In the former part of this paper, we explain a problem of p»n, which is the fundamental challenge in data analysis in bioinformatics. In particular, we consider a typical problem of p»n in prediction of treatment effects using microarray data as feature vectors. Then, we introduce some new boosting methods based on the area under the ROC curve. After showing some applications of the boosting methods, we summarize the present problems and refer to outlook for the future.