利用張量迴歸偵測基因與環境之交互作用

Abstract

偵測基因與環境之交互作用為生物醫學研究中重要的議題之一。雖然廣義線性模式已被廣泛地使用於解決這類型的問題，但將會遭遇高維度的困難。在此研究中，我們利用矩陣結構並應用張量迴歸模式的技術來克服偵測基因與環境之交互作用所遇到的高維度困難。張量迴歸的優點之一為，在模式建構上使用較精簡的參數個數。因此，我們期望利用張量迴歸能更有效地且更有力地來偵測基因與環境之交互作用。利用張量迴歸偵測基因與環境之交互作用的另一個優點為可同時估計effect sizes。我們藉由模擬分析以及兩個資料集來評估本研究提出之方法的表現。

Testing the significance of gene×environment (G×E) interactions is an important issue in biomedical research. Although the generalized linear model has been widely applied in this problem, it will suffer the difficulty of high-dimensionality. In this study, we utilize the matrix structure and apply the technique of tensor regression model to overcome the difficulty of high-dimensionality in detecting G×E. One advantage of tensor regression is the parsimony of parameters used. As a result, tensor regression is expected to be more efficient and powerful to detecting G×E. Another advantage of testing G×E by tensor regression is that the effect sizes can be estimated at the same time. We evaluate the performances of our methods through numerical studies and two data sets (the PSQI data and the EEG data).