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: | Charles Van De Mark |
| Institution: | Massachusetts Institute of Technology |
| Department: | Biological Engineering |
| Country: | |
| Proposed Analysis: | While many previous studies attempting a multi-omics profiling of AD have used cross-sectional data collected across several patients, we will use data from ADNI to study and forecast disease trajectory; this will uncover crucial insights for diagnostic and therapeutic strategies related to AD. The specific aims of our proposed project are: Aim 1: Develop a deep learning model to assess the risk of developing AD. We will integrate longitudinal neuroimaging and clinical data from ADNI to construct a predictive deep learning model, identify key prognostic and diagnostic features of AD and predict disease, and predict disease onset. Aim 2: Quantify AD progression and predict disease outcome. We will leverage multi-omics and neuroimaging data to develop a deep learning model that quantifies the progression of AD and forecasts disease outcome. This insight will not only be diagnostically useful, but it will also allow for the development of therapeutic intervention strategies specifically tailored to AD progression status. Aim 3: Identify critical biological features predisposing to and driving the progression of AD. In parallel to Aim 1 and 2, we will report biological insights drawn across neuroimaging, genetic, and clinical data. Such insights will be diagnostically and therapeutically relevant. Associations across data types are of particular interest. 1.1. Significance As there is currently an insufficient understanding of AD risk, onset, and progression, AD is especially difficult to diagnose and treat effectively. The integration of longitudinal multi-omics data sets with imaging data across a large cohort of individuals in a deep learning framework will make significant progress towards elucidating key molecular/imaging prognostic and diagnostic features associated with disease outcomes. This will ultimately facilitate highly accurate diagnosis of AD. Specifically, the models we develop will have three critical applications: the prediction of AD onset in individuals at risk for the disease, the diagnosis of individuals with early signs of AD, and the forecasting of disease progression in individuals once diagnosed with AD. The availability of longitudinal multi-omics and imaging data from ADNI will enable us to undertake this project. While our deep learning models will be impactful for their predictive prognostic and diagnostic utilities, the biological insights we will be able to draw may help uncover critical molecular and phenotypic features which may also then be used to diagnose and assess AD. The identification of imaging features associated with genetic features are of particularly well-suited to deep learning models. To summarize the significance of the project, critical genetic insights from neuroimaging data may facilitate rapid, accurate diagnosis of AD. Further, our findings may lead to novel therapeutic strategies for effective treatment of individuals with AD. 1.2. Innovation Previously, many multi-omics studies in Alzheimer’s disease primarily integrated imaging and clinical information including age, gender, and MMSE, without genetic information, where biological insights including the relationship between genetic and imaging patterns cannot be found. Alternatively, most imaging-genomics studies in Alzheimer’s disease mainly focused on genomics data, especially single nucleotide polymorphisms (SNPs) for either disease classification or progression analysis, which is “cross-sectional”. In the present study, we integrate multi-omics data with genetic and imaging data to perform AD risk prediction before disease onset and outcome prediction after disease onset, during which we will try to identify novel biomarkers and imaging markers that are linked to AD onset and progression. In summary, we will focus on multi-omics data before and after disease onset, where we will compare the imaging and expression profile of the patients at the two different stages. 3.3. Approach We plan to build separate models for imaging and genetic data before building the multi-omics model. We will use neural networks to build models for imaging data. To achieve interpretability, we will first use a fully convolutional network (FCN) to predict probability maps of outcome of interest to reduce dimensionality of data and obtain important features in imaging as described in (Qiu, et al., 2020). As for genetic data, we will use machine learning-based methods such as XGBoost and random forest to do the feature selection and model building. After the two models for imaging data and genetic data are ready, we will combine the selected features of both together with clinical information of the patients and use a multilayer perceptron (MLP) for outcome prediction. We will also build MLPs using clinical data only and imaging features only for comparison to see whether inclusion of other omics data improves the model performance. In this case, since we are interested in both risk of AD before disease onset and disease outcome after disease onset, we will use the abovementioned methods for both tasks except that response of the models are different. ROC-AUC, F1-score that considers both recall of a test and precision, and Matthew’s correlation will be used to evaluate model performance. After recording the results of these two tasks, we will perform downstream analysis including differential expression analysis, clustering, and building more models to see whether imaging features can be used to predict molecular features (or the inverse relationship). Qiu, S., Joshi, P. S., Miller, M. I., Xue, C., Zhou, X., Karjadi, C., ... & Kolachalama, V. B. (2020). Development and validation of an interpretable deep learning framework for Alzheimer’s disease classification. Brain, 143(6), 1920-1933. |
| Additional Investigators | |
| Investigator's Name: | Tong Ding |
| Proposed Analysis: | While many previous studies attempting a multi-omics profiling of AD have used cross-sectional data collected across several patients, we will use data from ADNI to study and forecast disease trajectory; this will uncover crucial insights for diagnostic and therapeutic strategies related to AD. The specific aims of our proposed project are: Aim 1: Develop a deep learning model to assess the risk of developing AD. We will integrate longitudinal neuroimaging and clinical data from ADNI to construct a predictive deep learning model, identify key prognostic and diagnostic features of AD and predict disease, and predict disease onset. Aim 2: Quantify AD progression and predict disease outcome. We will leverage multi-omics and neuroimaging data to develop a deep learning model that quantifies the progression of AD and forecasts disease outcome. This insight will not only be diagnostically useful, but it will also allow for the development of therapeutic intervention strategies specifically tailored to AD progression status. Aim 3: Identify critical biological features predisposing to and driving the progression of AD. In parallel to Aim 1 and 2, we will report biological insights drawn across neuroimaging, genetic, and clinical data. Such insights will be diagnostically and therapeutically relevant. Associations across data types are of particular interest. 1.1. Significance As there is currently an insufficient understanding of AD risk, onset, and progression, AD is especially difficult to diagnose and treat effectively. The integration of longitudinal multi-omics data sets with imaging data across a large cohort of individuals in a deep learning framework will make significant progress towards elucidating key molecular/imaging prognostic and diagnostic features associated with disease outcomes. This will ultimately facilitate highly accurate diagnosis of AD. Specifically, the models we develop will have three critical applications: the prediction of AD onset in individuals at risk for the disease, the diagnosis of individuals with early signs of AD, and the forecasting of disease progression in individuals once diagnosed with AD. The availability of longitudinal multi-omics and imaging data from ADNI will enable us to undertake this project. While our deep learning models will be impactful for their predictive prognostic and diagnostic utilities, the biological insights we will be able to draw may help uncover critical molecular and phenotypic features which may also then be used to diagnose and assess AD. The identification of imaging features associated with genetic features are of particularly well-suited to deep learning models. To summarize the significance of the project, critical genetic insights from neuroimaging data may facilitate rapid, accurate diagnosis of AD. Further, our findings may lead to novel therapeutic strategies for effective treatment of individuals with AD. 1.2. Innovation Previously, many multi-omics studies in Alzheimer’s disease primarily integrated imaging and clinical information including age, gender, and MMSE, without genetic information, where biological insights including the relationship between genetic and imaging patterns cannot be found. Alternatively, most imaging-genomics studies in Alzheimer’s disease mainly focused on genomics data, especially single nucleotide polymorphisms (SNPs) for either disease classification or progression analysis, which is “cross-sectional”. In the present study, we integrate multi-omics data with genetic and imaging data to perform AD risk prediction before disease onset and outcome prediction after disease onset, during which we will try to identify novel biomarkers and imaging markers that are linked to AD onset and progression. In summary, we will focus on multi-omics data before and after disease onset, where we will compare the imaging and expression profile of the patients at the two different stages. 3.3. Approach We plan to build separate models for imaging and genetic data before building the multi-omics model. We will use neural networks to build models for imaging data. To achieve interpretability, we will first use a fully convolutional network (FCN) to predict probability maps of outcome of interest to reduce dimensionality of data and obtain important features in imaging as described in (Qiu, et al., 2020). As for genetic data, we will use machine learning-based methods such as XGBoost and random forest to do the feature selection and model building. After the two models for imaging data and genetic data are ready, we will combine the selected features of both together with clinical information of the patients and use a multilayer perceptron (MLP) for outcome prediction. We will also build MLPs using clinical data only and imaging features only for comparison to see whether inclusion of other omics data improves the model performance. In this case, since we are interested in both risk of AD before disease onset and disease outcome after disease onset, we will use the abovementioned methods for both tasks except that response of the models are different. ROC-AUC, F1-score that considers both recall of a test and precision, and Matthew’s correlation will be used to evaluate model performance. After recording the results of these two tasks, we will perform downstream analysis including differential expression analysis, clustering, and building more models to see whether imaging features can be used to predict molecular features (or the inverse relationship). Qiu, S., Joshi, P. S., Miller, M. I., Xue, C., Zhou, X., Karjadi, C., ... & Kolachalama, V. B. (2020). Development and validation of an interpretable deep learning framework for Alzheimer’s disease classification. Brain, 143(6), 1920-1933. |
| Investigator's Name: | Ahmad Abdel-Azim |
| Proposed Analysis: | While many previous studies attempting a multi-omics profiling of AD have used cross-sectional data collected across several patients, we will use data from ADNI to study and forecast disease trajectory; this will uncover crucial insights for diagnostic and therapeutic strategies related to AD. The specific aims of our proposed project are: Aim 1: Develop a deep learning model to assess the risk of developing AD. We will integrate longitudinal neuroimaging and clinical data from ADNI to construct a predictive deep learning model, identify key prognostic and diagnostic features of AD and predict disease, and predict disease onset. Aim 2: Quantify AD progression and predict disease outcome. We will leverage multi-omics and neuroimaging data to develop a deep learning model that quantifies the progression of AD and forecasts disease outcome. This insight will not only be diagnostically useful, but it will also allow for the development of therapeutic intervention strategies specifically tailored to AD progression status. Aim 3: Identify critical biological features predisposing to and driving the progression of AD. In parallel to Aim 1 and 2, we will report biological insights drawn across neuroimaging, genetic, and clinical data. Such insights will be diagnostically and therapeutically relevant. Associations across data types are of particular interest. 1.1. Significance As there is currently an insufficient understanding of AD risk, onset, and progression, AD is especially difficult to diagnose and treat effectively. The integration of longitudinal multi-omics data sets with imaging data across a large cohort of individuals in a deep learning framework will make significant progress towards elucidating key molecular/imaging prognostic and diagnostic features associated with disease outcomes. This will ultimately facilitate highly accurate diagnosis of AD. Specifically, the models we develop will have three critical applications: the prediction of AD onset in individuals at risk for the disease, the diagnosis of individuals with early signs of AD, and the forecasting of disease progression in individuals once diagnosed with AD. The availability of longitudinal multi-omics and imaging data from ADNI will enable us to undertake this project. While our deep learning models will be impactful for their predictive prognostic and diagnostic utilities, the biological insights we will be able to draw may help uncover critical molecular and phenotypic features which may also then be used to diagnose and assess AD. The identification of imaging features associated with genetic features are of particularly well-suited to deep learning models. To summarize the significance of the project, critical genetic insights from neuroimaging data may facilitate rapid, accurate diagnosis of AD. Further, our findings may lead to novel therapeutic strategies for effective treatment of individuals with AD. 1.2. Innovation Previously, many multi-omics studies in Alzheimer’s disease primarily integrated imaging and clinical information including age, gender, and MMSE, without genetic information, where biological insights including the relationship between genetic and imaging patterns cannot be found. Alternatively, most imaging-genomics studies in Alzheimer’s disease mainly focused on genomics data, especially single nucleotide polymorphisms (SNPs) for either disease classification or progression analysis, which is “cross-sectional”. In the present study, we integrate multi-omics data with genetic and imaging data to perform AD risk prediction before disease onset and outcome prediction after disease onset, during which we will try to identify novel biomarkers and imaging markers that are linked to AD onset and progression. In summary, we will focus on multi-omics data before and after disease onset, where we will compare the imaging and expression profile of the patients at the two different stages. 3.3. Approach We plan to build separate models for imaging and genetic data before building the multi-omics model. We will use neural networks to build models for imaging data. To achieve interpretability, we will first use a fully convolutional network (FCN) to predict probability maps of outcome of interest to reduce dimensionality of data and obtain important features in imaging as described in (Qiu, et al., 2020). As for genetic data, we will use machine learning-based methods such as XGBoost and random forest to do the feature selection and model building. After the two models for imaging data and genetic data are ready, we will combine the selected features of both together with clinical information of the patients and use a multilayer perceptron (MLP) for outcome prediction. We will also build MLPs using clinical data only and imaging features only for comparison to see whether inclusion of other omics data improves the model performance. In this case, since we are interested in both risk of AD before disease onset and disease outcome after disease onset, we will use the abovementioned methods for both tasks except that response of the models are different. ROC-AUC, F1-score that considers both recall of a test and precision, and Matthew’s correlation will be used to evaluate model performance. After recording the results of these two tasks, we will perform downstream analysis including differential expression analysis, clustering, and building more models to see whether imaging features can be used to predict molecular features (or the inverse relationship). Qiu, S., Joshi, P. S., Miller, M. I., Xue, C., Zhou, X., Karjadi, C., ... & Kolachalama, V. B. (2020). Development and validation of an interpretable deep learning framework for Alzheimer’s disease classification. Brain, 143(6), 1920-1933. |
