A new framework of discrimination analysis for high-dimensional data based on estimation of data structure

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K12401
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Statistical science
• Research Institution: Nagoya University
• Principal Investigator: Matsui Shigeyuki 名古屋大学, 医学系研究科, 教授 (80305854)
• Co-Investigator(Kenkyū-buntansha): 小森 理 成蹊大学, 理工学部, 准教授 (60586379)
• Research Collaborator: Crowley John Chief of Strategic Alliances Cancer Research And Biostatistics, Board Chair
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 判別・予測解析 / 機械学習 / 疾患の異質性 / 統計モデリング

Outline of Final Research Achievements

We developed a novel framework of discrimination analysis of phenotype classes using high-dimensional genomic data in biomedical researches. This framework is based on hierarchical mixture models of the underlying structure on the association between the phonotype and genomic data and is expected to allow for stable discrimination and also for estimation of discrimination accuracy based on the model. We also considered incorporation of disease heterogeneity at the molecular level. One approach is the use of nested mixture models that can identify clusters of genes that are associated with the phonotype in particular subsets of disease patients. We applied the developed methods to real datasets from clinical genomic researches in cancer and other diseases.

Free Research Field

統計科学

Academic Significance and Societal Importance of the Research Achievements

ゲノムデータなどの多次元データを用いた提案する判別・予測解析は、ゲノムデータがもっている自然な関連構造、疾患の分子レベルでの異質性を明示的に考慮しており、統計・機械学習の新しい枠組みを提案するものである。一方で、本研究で開発した方法を適用することで、疾患の診断法の開発はもとより、疾患の分子機構の理解、新規治療法の分子標的の発見に役立つと期待できる。