A Study On Big Data-driven Biomarker Discovery

The thesis intends to address a few problems arouse in the field of big data-driven biomarker discovery, using a series of multi-scale mathematical modeling and statis-tical analysis. Over the past decade, biomedical studies have developed beyond the macroscopic level studies of the brain morphology and function and start to explore the microscopic level neural network as well as genetic pathways. Vast amount of data has been accumulated in related areas during this period. These multi-scale and multi-modal data calls for analytical tools to establish new biomarkers which can unveil the underlying biological mechanism. However, due to the diversity and comprehensive-ness of the related fields, there are a lot of complicated issues inevitably encountered during the study. Therefore it is essential to develop novel and efficient mathemati-cal and computational methods for seeking the integrated biomarkers across different levels and stages. To resolve the aforementioned challenges, we start by giving a brief introduction to the background knowledge regarding neuroimaging techniques, genom-ic analysis and imaging genetics. Then we use some mathematical models in exploring the biomarkers in attention deficit hyperactivity disorder (ADHD) disease, and the ef-fectiveness of our result is validated using pattern recognition study. To achieve an imaging-genetic analysis of ADHD disease, we propose a novel brain-wide association study (BWAS) under a general framework combining the power from genome-wide as-sociation studies (GWAS). By the integration of multiscale and multimodal statistical approaches, we present a comprehensive interpretation of the biomarkers in ADHD disease on neuroimaging and genomic levels. Meanwhile we conducted a robustness analysis on BWAS, and also benchmarked its performance and compared its statistical power with other alternative methods, such as the two-sample t-test, the permuta-tion test and logistic regression. Both the advantages and limitations of BWAS are discussed. In addition, given the important role fractional-order dynamical systems played in fractional diffusion process at molecular level upon which diffusion tensor imaging (DTI) technology was developed, we describe an approach for the identifi-cation of fractional-order systems (FOSs) with sparse structures in high dimensions, which will helps to investigate the underlying dynamical models through the large sample observed data. The thesis is concluded with some remarks and a discussion of future directions.