Integrating multi-omics datasets to infer phenotype-specific driver genes, regulatory interactions and drug response

点击翻译按钮获取中文摘要

PROJECT SUMMARY Alzheimer’s disease and its related dementias (ADRD) are a growing public health crisis with no known cure. Alzheimer’s disease (AD) is the most common cause of dementia, expected to affect about 50 million people globally by 2050. Despite extensive efforts, no effective treatment has been established. Recent efforts in drug development for AD have not advanced at the rate needed to discover new and diverse therapy modalities for the disease. Furthermore, due to complexity of AD pathogenesis and co- morbid conditions, combination drug therapy is essential for efficacious treatment for AD. As traditional drug discovery takes 10-15 years and costs $2-3 billion, computational drug discovery approaches have gained popularity in recent years. Particularly, repurposing existing approved drugs for new diseases could circumvent the labor- and cost-intensive process of traditional drug discovery. The aim of this project is to develop a novel machine learning method to integrate multi-modal drug, disease, and protein datasets to prioritize FDA-approved drugs and their combinations for AD/ADRD. To represent the multi- modal datasets and the multiple relationships between drugs, diseases, and targets, an attributed multiplex heterogeneous graph will be constructed. Inspired from growing research on deep graph representation learning methods, a novel deep learning architecture will be developed to learn node representations in this complex graph. Using these node representations, a machine learning model will be trained to rank drugs and their combinations based on their associations to AD and AD-related pathogenic processes and pathways. Using attention mechanisms, the proposed machine learning model will be interpretable. The expected outcome of this research is a ranked list of drug combinations for AD/ADRD, with novel drug-target interactions related to AD/ADRD along with key features and datasets that support these findings. The proposed project will complement the efforts of the parent award the goal of which is to develop open-source integrative computational tools that perform secondary analysis of publicly available multi-modal biological, clinical, and environmental exposure datasets to infer context-specific regulatory interactions and modules, and to predict disease associated genes and patient-specific drug response.