The etiology of many complex diseases, including Type-1 Diabetes (T1D), cannot be simply explained by genetic causes. Various factors, genetic as well as environmental, influence the progress of diseases. The critical issue to deriving the full benefit from biological, clinical, and longitudinal cohort studies for complex diseases is the appropriate analysis of the available large volumes of data, including these large-scale measurements and knowledge accrued from past research. Data mining approaches, especially feature selection from the massive number of measurements, become critical to identify reproducible and accurate risk factors to characterize pathogenic processes or pharmacologic responses to a therapeutic intervention for complex diseases including T1D. At the same time, data collection takes a significant amount of time and resources. Identifying risk factors and their interactions will significantly expedite the research at a low cost. The pri- mary objective of the proposed application is to develop a general network-based mathematical framework and efficient algorithms for identifying risk factors and their interactions as prognostic features that are highly informative about disease development. We will apply the developed algorithms to analyze the existing large-scale studies maintained at the Pediatric Epidemiology Center (PEC) at the University of South Florida (USF), including The Environmental Determinants of Diabetes in the Young (TEDDY) and the Diabetes Prevention Trial-Type 1 (DPT-1) studies. The identification of risk factors and their interactions provides deep insights to disease causality and mechanism. The proposed project has three specific aims: (1) An innovative data-driven analysis framework for risk factor identification will be presented and a general network-based mathematical model to identify risk factors and their interactions for disease development will be developed. (2) Fast and effective risk factor identification algorithms will be developed, which can be used to identify accurate synergistic factors. (3) The developed algorithms will be used for the large-scale studies, including TEDDY and DPT-1, to identify both genetic and environmental risk factors and their interactions with high predictive values for T1D development. We also will evaluate the performance of our algorithms in comparison with other traditional analysis for predicting the development and onset of T1D. Upon successful completion of this project, we expect that the developed algorithms will become a useful tool for biomedical data analysis with significant impacts on patient-oriented research for understanding the etiology, incidence, prevalence, natural history, and pathophysiology of T1D and other complex diseases. The proposed application will lay down the foundation and provide the direction for exploratory research on the re-use and analysis of existing data sets and the development of novel hypothesis and experiment design. 1

Public Health Relevance

This project aims to develop novel feature selection methods that can be practically employed in analyzing the existing large-scale studies, including The Environmental Determinants of Diabetes in the Young (TEDDY) and the Diabetes Pre- vention Trial-Type 1 (DPT-1) studies to aid in the identification and analysis of genetic as well as environmental risks, including demographic, dietary, immunologic, and metabolic markers, and their interactions for predicting progression to Type-1 Diabetes (T1D). Successful completion of the project will result in efficient feature selection algorithms for iden- tifying risk factors and understanding their interactions for the development of T1D, which will enable a paradigm shift from traditional hypothesis-driven analysis to data-driven analysis to generate working hypotheses to elucidate the etiology, incidence, prevalence, and pathophysiology of T1D and to design better screening strategies for early disease prediction and prevention. Although focusing on the existing TEDDY and DPT-1 studies in this project, the proposed feature selection methods are general and suitable for the analysis of large-scale studies for the identification of risk factors and their inter- actions for other complex diseases, which will serve as a solid foundation for our future data-driven projects to combine genomic, environmental, and clinical measurements from multiple studies maintained at the Pediatric Epidemiology Center (PEC) at the University of South Florida (USF), including, for example, DPT-1, TEDDY, TrialNET, TRIGR (Trial to Reduce Insulin-dependent diabetes mellitus in the Genetically at Risk), and RDCRN (Rare Diseases Clinical Research Network). 1

National Institute of Health (NIH)
National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)
Exploratory/Developmental Grants (R21)
Project #
Application #
Study Section
Biostatistical Methods and Research Design Study Section (BMRD)
Program Officer
Sechi, Salvatore
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of South Florida
Biostatistics & Other Math Sci
Schools of Engineering
United States
Zip Code
Wang, Yijie; Qian, Xiaoning (2016) Stochastic block coordinate Frank-Wolfe algorithm for large-scale biological network alignment. EURASIP J Bioinform Syst Biol 2016:9
Zamani Dadaneh, Siamak; Qian, Xiaoning (2016) Bayesian module identification from multiple noisy networks. EURASIP J Bioinform Syst Biol 2016:5
Lu, Meng; Lee, Hye-Seung; Hadley, David et al. (2014) Supervised categorical principal component analysis for genome-wide association analyses. BMC Genomics 15 Suppl 1:S10
Sajjadi, Seyed Javad; Qian, Xiaoning; Zeng, Bo et al. (2014) Network-Based Methods to Identify Highly Discriminating Subsets of Biomarkers. IEEE/ACM Trans Comput Biol Bioinform 11:1029-37
Lu, Meng; Lee, Hye-Seung; Hadley, David et al. (2014) Logistic Principal Component Analysis for Rare Variants in Gene-Environment Interaction Analysis. IEEE/ACM Trans Comput Biol Bioinform 11:1020-8
Wang, Yijie; Qian, Xiaoning (2014) Joint clustering of protein interaction networks through Markov random walk. BMC Syst Biol 8 Suppl 1:S9
Wang, Yijie; Qian, Xiaoning (2014) Functional module identification in protein interaction networks by interaction patterns. Bioinformatics 30:81-93
Ren, Shaogang; Zeng, Bo; Qian, Xiaoning (2013) Adaptive bi-level programming for optimal gene knockouts for targeted overproduction under phenotypic constraints. BMC Bioinformatics 14 Suppl 2:S17
Wang, Yijie; Qian, Xiaoning (2013) A novel subgradient-based optimization algorithm for blockmodel functional module identification. BMC Bioinformatics 14 Suppl 2:S23