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: | Jordi Abante Llenas |
| Institution: | Universitat de Barcelona |
| Department: | School of Medicine |
| Country: | |
| Proposed Analysis: | Traditional GWAS analysis obviates interactions between single nucleotide polymorphisms (SNPs), ignoring epistatic interactions that could be critical for our biological understanding and drug development. This is because the amount of data required to detect interactions between SNPs in the commonly used logistic regression models would be prohibitive. However, the study of epistatic interactions in a highly complex disease such as AD can reveal new disease modifiers, and novel avenues for drug development since targetable genes that are critical for the disease progression might only be relevant through interactions. To address this problem, we plan to leverage recent advances in graph neural networks (GNNs), which have shown remarkable performance in other applications. By encoding information about biological processes in graphs that establish biologically meaningful relationships between SNPs, we plan to identify informative processes in distinguishing healthy from AD-affected individuals and predicting the onset of the disease. Using prior knowledge about biological processes will narrow down the search space and will allow us to perform inference on the graphs using deep learning techniques, modeling non-linear effects commonly missed by traditional linear models. Indeed, recent work by Hernández-Lorenzo et al. 2022 was the first to apply GNNs to the study of AD, showing great promise. We will build on their work by extending the network beyond the few genes and SNPs used in their analysis. To that end, we will include all available variants, including variants in non-coding regions, and encode SNPs in a way that allows us to consider epistatic interactions. Furthermore, given that we plan to build a larger graph, we will include attention mechanisms to the network for improved interpretability and identify the critical subnetworks. Finally, we also plan to use semi-supervised learning techniques to discover potential novel interactions not represented in the protein-protein interaction networks. We believe that the power of GNNs can help us better understand and predict a highly complex disease such as AD. If successful, this work can lead to clinically relevant results for the early diagnosis of AD. |
| Additional Investigators | |
| Investigator's Name: | Riccardo Smeriglio |
| Proposed Analysis: | Traditional GWAS analysis obviates interactions between single nucleotide polymorphisms (SNPs), ignoring epistatic interactions that could be critical for our biological understanding and drug development. This is because the amount of data required to detect interactions between SNPs in the commonly used logistic regression models would be prohibitive. However, the study of epistatic interactions in a highly complex disease such as AD can reveal new disease modifiers, and novel avenues for drug development since targetable genes that are critical for the disease progression might only be relevant through interactions. To address this problem, we plan to leverage recent advances in graph neural networks (GNNs), which have shown remarkable performance in other applications. By encoding information about biological processes in graphs that establish biologically meaningful relationships between SNPs, we plan to identify informative processes in distinguishing healthy from AD-affected individuals and predicting the onset of the disease. Using prior knowledge about biological processes will narrow down the search space and will allow us to perform inference on the graphs using deep learning techniques, modeling non-linear effects commonly missed by traditional linear models. Indeed, recent work by Hernández-Lorenzo et al. 2022 was the first to apply GNNs to the study of AD, showing great promise. We will build on their work by extending the network beyond the few genes and SNPs used in their analysis. To that end, we will include all available variants, including variants in non-coding regions, and encode SNPs in a way that allows us to consider epistatic interactions. Furthermore, given that we plan to build a larger graph, we will include attention mechanisms to the network for improved interpretability and identify the critical subnetworks. Finally, we also plan to use semi-supervised learning techniques to discover potential novel interactions not represented in the protein-protein interaction networks. We believe that the power of GNNs can help us better understand and predict a highly complex disease such as AD. If successful, this work can lead to clinically relevant results for the early diagnosis of AD. |
| Investigator's Name: | Joana Rosell |
| Proposed Analysis: | Traditional GWAS analysis obviates interactions between single nucleotide polymorphisms (SNPs), ignoring epistatic interactions that could be critical for our biological understanding and drug development. This is because the amount of data required to detect interactions between SNPs in the commonly used logistic regression models would be prohibitive. However, the study of epistatic interactions in a highly complex disease such as AD can reveal new disease modifiers, and novel avenues for drug development since targetable genes that are critical for the disease progression might only be relevant through interactions. To address this problem, we plan to leverage recent advances in graph neural networks (GNNs), which have shown remarkable performance in other applications. By encoding information about biological processes in graphs that establish biologically meaningful relationships between SNPs, we plan to identify informative processes in distinguishing healthy from AD-affected individuals and predicting the onset of the disease. Using prior knowledge about biological processes will narrow down the search space and will allow us to perform inference on the graphs using deep learning techniques, modeling non-linear effects commonly missed by traditional linear models. Indeed, recent work by Hernández-Lorenzo et al. 2022 was the first to apply GNNs to the study of AD, showing great promise. We will build on their work by extending the network beyond the few genes and SNPs used in their analysis. To that end, we will include all available variants, including variants in non-coding regions, and encode SNPs in a way that allows us to consider epistatic interactions. Furthermore, given that we plan to build a larger graph, we will include attention mechanisms to the network for improved interpretability and identify the critical subnetworks. Finally, we also plan to use semi-supervised learning techniques to discover potential novel interactions not represented in the protein-protein interaction networks. We believe that the power of GNNs can help us better understand and predict a highly complex disease such as AD. If successful, this work can lead to clinically relevant results for the early diagnosis of AD. |
