There are many active research projects accessing and applying shared ADNI data. Use the search above to find specific research focuses on the active ADNI investigations. This information is requested annually as a requirement for data access.
| Principal Investigator | |
| Principal Investigator's Name: | Jeffrey Brabec |
| Institution: | University of Vermont |
| Department: | Neurological Sciences |
| Country: | |
| Proposed Analysis: | The goal of this project is to identify novel tissue-specific Alzheimer’s Disease (AD) risk gene networks through systems genetics analysis of imaging data. In recent years, researchers have improved systems biological analysis complex diseases in two important directions. First, network-based machine learning tools, such as NetWAS, use functional interaction data to identify novel risk genes from GWAS data that fall below classical statistical thresholds, but nevertheless confer disease risk. Second, computer vision technologies have defined novel imaging phenotypes that predict diagnosis rigorously and robustly. In the context of AD, NetWAS has suggested a novel association between protocadherins and a hippocampal volume phenotype. Likewise, convolutional neural networks (CNNs) can diagnose AD using radiographic imaging. In making diagnoses, CNNs define novel radiographic features that discern cognitively normal individuals from those suffering from more advanced stages of AD. These features represent a novel class of endophenotypes whose genetics is largely unexplored. However, these tools have not been integrated together, and there are important outstanding conceptual hurdles to overcome. We propose to optimize NetWAS for identifying disease-relevant SNPs driving variation in both CNN-derived and psychological AD traits. We have three distinct aims with the ADNI data. First, we would like to optimize the choice of tissue network for NetWAS, as many complex neurological traits are distributed over many brain regions and tissue types. We will recapitulate previous NetWAS analyses of hippocampal and amygdalar volumes (Song et al., 2016, Yao et al., 2017), and show that NetWAS is most sensitive and specific for those tissue networks compared to alternative brain region and cell type networks. This will demonstrate that sensitivity and specificity of NetWAS results in a given tissue is a signal that that tissue is relevant to the analyzed trait. We will then apply NetWAS to the results of GWAS scans in the subsequent aims. Next, we will perform GWAS analysis of convolutional neural network (CNN) phenotypes of ADNI T2 weighted hippocampal MRI images. Liu et al. (2019) recently designed a CNN that predicts AD diagnosis, taking into account a patient’s age and natural age-related brain-tissue shrinkage. While their goal was to automate diagnosis and identify novel diagnostic features, our goal is to identify the genetic influences on these novel features. We expect to identify novel, significant SNPs for these imaging endophenotypes, and see an improved p-value for known AD-related SNPs, demonstrating the increased mapping power of these traits over categorial diagnosis. Finally, we would like to estimate the SNP-based heritability and compute polygenic risk scores for CNN-derived traits. We will perform regularized canonical correlation analysis of genotype and phenotype, as has recently been analyzed by Mitteroecker et al. (2016) in the context of quantitative genetics and applied to imaging genetics in AD by Lorenzi et al. (2018). We expect CNN-based imaging traits to be more heritable (and the corresponding risk scores more robust) than categorical diagnosis. |
| Additional Investigators | |
| Investigator's Name: | Matt Mahoney |
| Proposed Analysis: | The goal of this project is to identify novel tissue-specific Alzheimer’s Disease (AD) risk gene networks through systems genetics analysis of imaging data. In recent years, researchers have improved systems biological analysis complex diseases in two important directions. First, network-based machine learning tools, such as NetWAS, use functional interaction data to identify novel risk genes from GWAS data that fall below classical statistical thresholds, but nevertheless confer disease risk. Second, computer vision technologies have defined novel imaging phenotypes that predict diagnosis rigorously and robustly. In the context of AD, NetWAS has suggested a novel association between protocadherins and a hippocampal volume phenotype. Likewise, convolutional neural networks (CNNs) can diagnose AD using radiographic imaging. In making diagnoses, CNNs define novel radiographic features that discern cognitively normal individuals from those suffering from more advanced stages of AD. These features represent a novel class of endophenotypes whose genetics is largely unexplored. However, these tools have not been integrated together, and there are important outstanding conceptual hurdles to overcome. We propose to optimize NetWAS for identifying disease-relevant SNPs driving variation in both CNN-derived and psychological AD traits. We have three distinct aims with the ADNI data. First, we would like to optimize the choice of tissue network for NetWAS, as many complex neurological traits are distributed over many brain regions and tissue types. We will recapitulate previous NetWAS analyses of hippocampal and amygdalar volumes (Song et al., 2016, Yao et al., 2017), and show that NetWAS is most sensitive and specific for those tissue networks compared to alternative brain region and cell type networks. This will demonstrate that sensitivity and specificity of NetWAS results in a given tissue is a signal that that tissue is relevant to the analyzed trait. We will then apply NetWAS to the results of GWAS scans in the subsequent aims. Next, we will perform GWAS analysis of convolutional neural network (CNN) phenotypes of ADNI T2 weighted hippocampal MRI images. Liu et al. (2019) recently designed a CNN that predicts AD diagnosis, taking into account a patient’s age and natural age-related brain-tissue shrinkage. While their goal was to automate diagnosis and identify novel diagnostic features, our goal is to identify the genetic influences on these novel features. We expect to identify novel, significant SNPs for these imaging endophenotypes, and see an improved p-value for known AD-related SNPs, demonstrating the increased mapping power of these traits over categorial diagnosis. Finally, we would like to estimate the SNP-based heritability and compute polygenic risk scores for CNN-derived traits. We will perform regularized canonical correlation analysis of genotype and phenotype, as has recently been analyzed by Mitteroecker et al. (2016) in the context of quantitative genetics and applied to imaging genetics in AD by Lorenzi et al. (2018). We expect CNN-based imaging traits to be more heritable (and the corresponding risk scores more robust) than categorical diagnosis. |
