Tingting Zhang, Ph.D. - US grants

Rapid and accurate detection of biothreat is important not only for containing its potential damages, but also for determining potential medical remedies. Extensive researches show that certain genes in infected cells have different mRNA expression levels for different pathogens.
Thus, an accurate identification of the genes that react to pathogens and an accurate quantification of their expression variations are key steps in early biothreat detections. The emerging RNA-Seq technologies provide tens of millions of short sequence reads of the expressed genes, which, after mapping to the genome, can be converted to accurately represent gene expression levels. However, the conversion from sequence reads to gene expression levels is still problematic. In this project, The investigator and her colleagues will tackle this problem by modeling RNA-Seq data through a broad class of flexible nonlinear models, called sufficient dimension reduction (SDR) models; propose novel variable selection methods for SDR models; and develop theoretical underpinning of the effectiveness of the proposed methods. As a consequence, this effort will result in a powerful software suite for estimating gene expression levels from RNA-seq data and identifying marker genes reacting to specific pathogens in a unified framework.


This project not only addresses some emerging issues in biothreat detections using high-throughput sequencing technologies, but also results in novel statistical methods and
theory broadly applicable to general statistical learning and prediction problems. More specifically, the proposed methods (i) produce innovative new methodologies for analyzing ultra-high dimensional data, (ii) inspire new lines of quantitative investigations in genomics, and (iii) offer a unique educational experience for both undergraduate and graduate students to participate in cutting-edge statistical and interdisciplinary research.

This project consists of two components, each motivated by the inference problem for functional magnetic resonance imaging (fMRI) data. In the first part, within the framework of generalized functional linear model (GFLM), a flexible semi-parametric model for neural hemodynamic response in the form of slope functions is introduced. To accommodate the variation of brain activity across different regions, stimulus types, and subjects, the new approach assumes the slope functions share the same but unknown functional shape for a given region and stimulus, while having subject-specific height, time to peak, and width. Several fast algorithms based on B-spline smoothing are proposed to estimate the model parameters for whole-brain analysis. The second part of the research focuses on building a novel Bayesian variable selection framework to study the relationship between individual traits and brain activity. The spline estimates of the brain hemodynamic responses from the first part are taken as predictors in a regression model where the response is the individual traits. Two types of priors are introduced jointly to achieve simultaneous variable selection and clustering.

FMRI is one of the most effective neuroimaging technologies for understanding brain activity. In recent years, fMRI data collected from complex studies with multiple subjects have been widely used in psychological and medical research. This project will provide tools for modeling, analysis and computation for this type of fMRI data. Project findings will advance basic understanding of the inter-relations between nature and nurture in shaping individual differences in brain function and behavior, and suggest new directions for interdisciplinary research that combines statistics, neuroscience and psychology. The open source R/Matlab software developed from the research will provide valuable data analysis and educational tools for the scientific community.

This National Science Foundation award provides support for the CBMS Conference: Elastic Functional and Shape Data Analysis (EFSDA), which will be held in July 16-20, 2018 at The Ohio State University in Columbus, OH. The primary lecturer is Professor Anuj Srivastava from the Department of Statistics at Florida State University. The conference will feature a lecture series on elastic methods for statistical analysis of functional and shape data, using tools from Riemannian geometry, Hilbert space methods, and computational science. The main focus of this conference will be on geometric approaches, especially on using elastic Riemannian metrics with desired invariance properties, and square-root representations of shape that have proven to simplify computations. These approaches enable joint registration and statistical analysis of functional data, and are termed elastic for that reason. The statistical goals include comparisons, summarization, clustering, modeling, and testing of functional and shape data objects. The proposed tools for statistical analysis of functional and shape data have broad applications in almost all branches of science. Any promotion of education, training, and collaboration in this cutting-edge research area will have a strong impact on the community. The audience for this workshop will include early career researchers from statistics, applied mathematics, engineering, computer science and biological sciences. By training and educating researchers in an important STEM area, this effort will facilitate future interdisciplinary collaborations amongst participants.

Recent years have seen a tremendous advancement in the use of Riemannian geometry in statistical data analysis, especially in shape analysis. EFSDA brings together tools from diverse disciplines, such as geometry, statistics, functional data analysis, computational science, and application domains, to develop a broad and comprehensive package of solutions. On one hand, it poses fundamental mathematical questions, including existence and uniqueness of optimal functional matching, and on the other, it provides efficient computational implementations for problems dealing with alignment, dimension reduction, and statistical modeling of functional and shape data. Given the proliferation of functional data in all scientific disciplines, these tools will help address important and urgent data analysis needs. The classroom-style lectures will be enhanced by multiple discussion sessions. For the conference webpage, please see https://stat.osu.edu/cbms-efsda.

This interdisciplinary research project will develop new statistical methods to analyze multi-subject, stimulus-evoked functional magnetic resonance imaging (fMRI) data collected from a psychology study of human emotion. The project will increase understanding of how human brain circuits associated with emotions function in a context that combines social support with externally generated emotional stress. Ultimately, the project will contribute knowledge of how the brain uses social support via the social regulation of emotion. This knowledge will facilitate future research in this area. The results of this research will assist clinical researchers interested in the neuropathology of many neurodevelopmental and affective disorders affecting children and adults. The project will provide the opportunity for undergraduate and graduate students (especially those from underrepresented groups) to participate in advanced statistical and multidisciplinary research involving human brain data. Project results, including scientific findings and developed software, will be made publicly available using public repositories.

The statistical models and computational methods to be developed will address typical challenges in analyzing fMRI data, including massive data size, complex spatial and temporal properties, and a weak signal-to-noise ratio. The new low-rank multivariate general linear models for multi-subject, stimulus-evoked fMRI data feature the brain activity's common properties shared across different regions, subjects, and stimulus types, and they require fewer parameters than nonparametric methods to characterize variation in brain activity. As such, the new approaches to fMRI data analysis are characterized by simultaneously reduced model parameters, increased estimation efficiency, and sufficient model flexibility. This project will develop new nonconvex optimization algorithms to address the computational challenges in analyzing fMRI data. Applying the developed methods to a fMRI study of human emotion, the investigators will examine the difference in brain responses to negative emotional stimuli under different social contact conditions and identify the association between emotion-related brain functions and concomitant affective feelings under different social contact conditions.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.