Grant: Canada Research Chairs

Data analyses are broadly impacted by variations in software and hardware processing environments, which questions the accuracy, robustness, and reproducibility of derived predictions in a number of data science fields. In neuroscience, studies showed that processing a given data set with different analytical approaches may result in conflicting conclusions, which threatens the trustworthiness of data analyses. Thus, analytical variability is an important issue that needs to be fully understood and efficiently addressed. The long-term goal of my research program is to support the reliability, robustness, and precision of data analyses by investigating analytical variability. In the coming years, I will apply my methodologies to the analysis of brain Magnetic Resonance Imaging (MRI) data for Parkinson's disease, a challenging field for which analytical variability is particularly crucial. My research program adopts an integrative, multi-disciplinary approach that investigates analytical variability from its numerical foundations to its data-science aspects up to its clinical materializations. The specific objectives of my research program are (1) to conduct exploratory evaluations of analytical variability: building on tools developed during my previous term as Canada Research Chair (Tier II), I will explore and characterize the range of analytical variability patterns that are currently impacting data science; (2) to leverage analytical variability as a tool: I will design methods to combine results obtained with different analytical approaches into more reliable ones, using approaches similar to ensemble learning; and (3) to improve the reliability of MRI measures of Parkinson’s disease through analytical variability: I will apply the knowledge and methods produced by (1) and (2) to the specific case of MRI measures of Parkinson's disease. My research program will demystify analytical variability for widely-used data analyses, it will produce open-source tools to support the investigation of analytical variability at various levels, and it will identify replicable, more robust MRI measures of Parkinson's disease. Throughout this program, I will train graduate students in data science, machine learning, numerical analysis, medical image analysis, software development, and high-performance computing, all fields in high-demand in industry and academia.