College of Information Engineering, Shanghai Maritime University, 1550 Haigang Ave, Shanghai, 201306, People's Republic of China.
J Mol Neurosci. 2022 Aug;72(8):1749-1763. doi: 10.1007/s12031-022-02031-9. Epub 2022 Jun 13.
Imaging genetics using imaging technology is regarded as a neuroanatomical phenotype to evaluate gene single nucleotide polymorphisms and their effects on the structure and function of different brain regions. It plays a vital role in bridging the initial understanding of the genetic basis of brain structure and dysfunction. Sparse canonical correlation analysis (SCCA) has become a widespread technique in this field because of its powerful ability to identify bivariate relationships and feature selection. Since most traditional SCCA algorithms assume that the input features are independent, this method obviously cannot be used to analyze genetic image data. The MT-SCCA model is unsupervised and cannot identify the genotype-phenotype associations for diagnostic guidance. Meanwhile, a single biological clinical index cannot fully reflect the physiological process of a comprehensive disease. Therefore, it is necessary to find biomarkers that can reflect Alzheimer's disease and physiological functions that can more comprehensively reflect the development of the disease. This article uses a multi-task sparse canonical correlation analysis and regression (MT-SCCAR) model to combine the annual depression level total score (GDSCALE), clinical dementia assessment scale (GLOBAL CDR), functional activity questionnaire (FAQ), and neuropsychiatric Symptom Questionnaire (NPI-Q) in this paper. These four clinical data are used as compensation information and embedded in the algorithm in a linear regression manner. It also reflects its superiority and robustness compared to traditional correlation analysis methods on actual and simulated data. Meanwhile, compared with MT-SCCA, the model utilized in this paper obtains a higher gene-ROI weight and identifies clearer biomarkers, which provides a practical basis for the study of complex human disease pathology.
使用成像技术进行影像遗传学研究被视为一种神经解剖学表型，用于评估基因单核苷酸多态性及其对不同脑区结构和功能的影响。它在连接对大脑结构和功能障碍的遗传基础的初步理解方面发挥着至关重要的作用。稀疏典型相关分析（SCCA）已成为该领域广泛使用的技术，因为它具有识别双变量关系和特征选择的强大能力。由于大多数传统的 SCCA 算法假设输入特征是独立的，因此该方法显然不能用于分析遗传图像数据。MT-SCCA 模型是无监督的，无法识别基因型-表型关联以进行诊断指导。同时，单一的生物学临床指标不能充分反映全面疾病的生理过程。因此，有必要寻找能够反映阿尔茨海默病的生物标志物和能够更全面反映疾病发展的生理功能。本文使用多任务稀疏典型相关分析和回归（MT-SCCAR）模型，将年度抑郁水平总分（GDSCALE）、临床痴呆评估量表（GLOBAL CDR）、功能活动问卷（FAQ）和神经精神症状问卷（NPI-Q）这四个临床数据结合起来。这些四个临床数据被用作补偿信息，并以线性回归的方式嵌入到算法中。它还反映了其在实际和模拟数据上相对于传统相关分析方法的优越性和稳健性。同时，与 MT-SCCA 相比，本文所采用的模型获得了更高的基因-ROI 权重，并识别出更清晰的生物标志物，为研究复杂人类疾病病理学提供了实际依据。
