Analytic Methods for Determining Multimodal Biomarkers for Parkinson's Disease (U18)
There are many exciting technological advances that provide massive amounts of medical information. Despite these breakthroughs, however, a considerable amount of this newly available information is underutilized. Better methods for using different types of information are needed to improve our ability to discover more, and to discover things faster. This is true for many scientific areas, including Parkinson's disease biomarkers research. Dr. Bowman and his team are experts in statistics, a scientific approach, in which they use mathematics as analysis tools to help them evaluate many different types of large datasets not commonly analyzed together. They are going to use this project to develop tools which will allow analysis of clinical, imaging, genetic, and biological information from subjects in the Kaiser-Permanente database and from the Emory University Morris K Udall Center of Excellence in Parkinson's disease. These tools will allow other researchers to also examine data that they collect via the NINDS Parkinson's Disease Biomarkers Program. In this way, biomarkers can be discovered that combine clinical, imaging, genetic, or other biological information, to create a constellation of biomarkers that could be useful in clinical trials and in the clinic. The methods Dr. Bowman and his team will develop will be made available to the scientific community as "freeware."
There has been considerable progress in understanding the biology of Parkinson's disease (PD). Reliable biomarkers are still lacking, however, for early stage detection of PD and for characterizing disease progression. Advances in biotechnology have led to the advent of mental health studies that collect large-scale, multi-dimensional data sets, including brain imaging data, genomic data, and biologic and clinical measures. Such studies provide an unprecedented opportunity for cross-cutting investigations that stand to gain a deeper understanding of PD. A major limiting factor to multidimensional biomarker development, however, is the lack of statistical tools available to accommodate diverse, large-scale data. Leveraging data from neuromelanin magnetic resonance imaging (NM-MRI) of the locus coeruleus and the substantia nigra, chemical shift imaging (CSI), diffusion tensor imaging (DTI), resting-state functional MRI, cerebrospinal fluid (CSF) analytes, genotype information, and numerous clinical variables, we plan to develop novel statistical techniques to identify multimodal PD biomarkers. Our data provide an unprecedented opportunity for cross-cutting methodological advances in multimodal PD biomarker discovery. Separately, we will consider a massive patient database with nearly 250,000 subscribers in Georgia. Building on our collective expertise in developing statistical and machine-learning methods for large-scale imaging data and in the pathophysiology of PD, we plan to advance methods for PD biomarker analyses and discovery through the following specific aims. First, we plan to develop new statistical techniques to reveal multimodal biomarkers for PD including imaging, clinical, and biologic variables. Secondly, we plan to utilize the massive clinical database to identify clinical risk factors for early stage PD. Thirdly, we will develop software equipped with a friendly graphical user interface (GUI) to implement the multimodal biomarker detection methods.
Goals of Project
- Resource Building: Develop new statistical techniques to reveal multimodal biomarkers for PD including imaging, clinical, and biologic variables. Challenge: To extract modality-specific and multimodal biomarkers for PD from possibly hundreds of thousands of variables combined from multiple neuroimaging modalities, genetic, molecular, and clinical information.
- Discovery: Identify clinical risk factors for early stage PD from a massive patient database. To identify early risk factors for PD from a massive patient database with hundreds of variables including, diagnostic information (e.g. depression, REM behavior disorder), medication history (e.g. constipation medicines, antidepressants), and lab results. We will use state-of-the-art statistical variable selection and regularization methods to determine important clinical variables preceding a PD diagnosis and will use these identified factors to quantify an individualized risk score reflecting the probability of progression to PD.
- Resource Building: Develop software equipped with a friendly graphical user interface (GUI) to implement the multimodal biomarker detection methods. To make our advanced statistical and machine-learning techniques readily available to other PD researchers. We will create a GUI that will facilitate the processes for reading in and mining large scale multimodal data for PD biomarker discovery.