Generalized Partially Additive Models For High-Dimensional Data

研究人员研究了广义可加部分线性模型(GAPLM)，目的是开发有效和灵活的估计和推断方法、变量选择程序、模型规格检验和模型结构检验，并研究这些方法在生物医学研究中的应用。具体地说，他(A)正在开发一种真正的方法，该方法能够选择重要的参数和非参数组件，即使在非参数组件和参数组件的数量不同的情况下也是如此；(B)正在开发GAPLM的模型规格测试；(C)研究GAPLM的模型结构确定；(D)研究相关数据的边际GAPLM；以及(E)应用先进的模型和建议的方法来分析基因数据，以研究某些疾病和基因之间的关系，包括识别癌细胞对不同药物治疗的特征基因表达谱，通过整合来自基因组和基因组标记的遗传变异的知识来预测遗传风险，以及从通过下一代测序平台收集的数据验证表观遗传密码。该项目的结果可以帮助识别重要的基因表达谱和癌细胞，并在癌症研究中追踪疾病进展。理论结果有助于发展变量选择的统计理论和高维协变量的半参数推断。

The investigator studies generalized additive partially linear models (GAPLM) with the aim of developing efficient and flexible estimation and inference methods, variable selection procedures,model specification tests, and model structure checks, and studies applications of these methods for biomedical research. Specifically speaking, he (a) is developing a genuine method that is able to select important parametric and nonparametric components that are numerically stable , even when the numbers of the nonparametric and parametric components diverge; (b) is developing model specification tests for GAPLM; (c) studies model structure determination for GAPLM; (d) studies marginal GAPLM for correlated data; and (e) applies the advanced models and proposed methods to analyze gene data for study of the relationship between certain diseases and genes, including the identification of signature gene expression profiles of cancer cells in response to different drug treatments, the prediction of genetic risks through integrating knowledge from genetic variations in the genome and genomic markers, and validation of epigenetic codes from data collected through next generation sequencing platforms.The proposed models and methods are motivated by the investigator's study of gene and other potentially useful biomarkers in cancer clinical trials. The results of this project can help identify important gene expression profiles and cancer cells and trace the disease progression in cancer research. The theoretic results contribute to the advancement of the statistical theory on variable selections and semi-parametric inference with high-dimensional covariates.