This project, Empowering Personalized Medicine: Integrating Imaging, Genetics and Biomarkers, responds to RFA-MH-12-020, entitled Integrating Multi-Dimensional Data to Explore Mechanisms Underlying Mental Disorders. By bringing together experts in neuroimaging, genetics, and mathematics, we plan to create an advanced, portable framework to combine diverse biomedical data from 3D neuroimaging (MRI, amyloid/FDG-PET), gene expression networks, genome-wide association studies (GWAS), and other multidimensional data (e.g., physiological biomarkers, epigenetic data, etc.). Our overall goal is to improve diagnosis and prognosis of disease by combining multiple levels of biological information (personalized medicine). In doing so, novel mathematical tools will automatically discover which biomarkers are most helpful in different contexts. To discover and test relationships between very high-dimensional measures (such as images and genomes), we use novel concepts for data reduction such as penalized regression (elastic nets), adaptive hierarchical clustering, Bayesian networks, and support vector machines. Avoiding the limitations of current work that tests individual gene effects independently, we extend the analysis of gene expression networks to images, to relate signs of disease to their genetic underpinnings and to all available biomarkers.
Aim 1 empowers discovery genetic variants (identified in GWAS, whole-exome and whole-genome sequencing) that modulate measures of disease. We will use compressive coding models to discover and verify which sets of genetic variants affect multidimensional images (e.g., co-registered MRI &PET, DTI). We will verify our predictions using k-fold cross-validation and independent replications in new samples and controllable test data.
Aim 2 extends our work using weighted gene co-expression network analysis (WGCNA) from single traits to entire databases of 3D images (MRI/PET). Our framework will merge GWAS, eQTL analysis, and expression-phenotype analysis but will be broadly applicable to any future high-throughput biological information (e.g. methylation profiles, DTI, fMRI).
In Aim 3, we will quantify the added predictive value derivable from genotyping, gene expression profiling, and multimodal neuroimaging for personalized prognosis and diagnosis. For example, which biomarkers (gene expression, CSF, MRI) are most useful in which cases? To maximize impact of this effort, we and our collaborators will test our tools on existing and new datasets from a range of neuropsychiatric disorders including frontotemporal dementia, Alzheimer's disease, schizophrenia, bipolar disorder, and autism (see Support Letters). All tools will be disseminated and linked to web-accessible databases that store and ease access to high-throughput genetic, genomic, and imaging datasets.
Our project improves diagnosis and predictions of patient outcomes by integrating diverse types of patient data including neuroimaging, gene expression profiles, cognitive, and behavioral markers of disease diagnosis, progression, and treatment response. To tackle these complex data types, we develop novel machine learning, network analysis, and database methods. Our personalized medicine approach will help researchers study and evaluate neurological and psychiatric conditions such as Alzheimer's disease, frontotemporal dementia, schizophrenia, bipolar disorder, and autism.
Showing the most recent 10 out of 62 publications
