Motivation: The molecular mechanism of Alzheimer's disease (AD) has not been clearly revealed and there is no clinically reliable genetic risk factor. Therefore, diagnosis of AD has been mostly performed by analyzing brain images such as magnetic resonance imaging and neuropsychological tests. Identifying the molecular-level mechanism of AD has been lacking data owing to the difficulty of sampling in the posterior brains of normal and AD patients; however, recent studies have produced and analyzed large-scale omics data for brain areas such as prefrontal cortex. Therefore, it is necessary to develop AD diagnosis or prediction methods based on these data. Results: This paper proposed a deep learning-based model that can predict AD using large-scale gene expression and DNA methylation data. The most challenging problem in constructing a model to diagnose AD based on the multi-omics dataset is how to integrate different omics data and how to deal with high-dimensional and low-sample-size data. To solve this problem, we proposed a novel but simple approach to reduce the number of features based on a differentially expressed gene and a differentially methylated position in the multi-omics dataset. Moreover, we developed a deep neural network-based prediction model that improves performance compared to that of conventional machine learning algorithms. The feature selection method and the prediction model presented in this paper outperformed conventional machine learning algorithms, which utilize typical dimension reduction methods. In addition, we demonstrated that integrating gene expression and DNA methylation data could improve the prediction accuracy. Availability: https://github.com/ChihyunPark/DNN_for_ADprediction.
All Science Journal Classification (ASJC) codes
- Computer Science Applications
- Artificial Intelligence