Networks are widely used in molecular biology to model interactions among components of biological systems and gain insight into changes in biological mechanism associated with various diseases. Recent evidence suggests that changes in biological networks, including rewiring or disruption of key interactions, may be associated with the development of complex diseases. This research project is motivated by the study of changes in metabolic networks in association with evolution and aging. The project aims to develop new statistical machine learning methods to determine whether evolutionary changes are manifested through changes in how metabolites interact with each other in metabolic pathways. More broadly, the methodologies under development in this project and the accompanying software will provide novel tools for biomedical researchers to infer differential patterns of connectivity in molecular networks associated with complex diseases.
The methodologies under development in this project utilize the framework of graphical models to estimate interactions among components of biological networks and identify changes in such networks associated with evolution and aging. Probabilistic graphical models provide a general framework for modeling interactions among random variables. While recent methodological and theoretical advances have facilitated the applications of graphical models to analysis of high-dimensional biological networks, existing methods are not applicable to heterogeneous and non-Gaussian observations obtained from mass-spectrometry-based metabolomics profiling experiments in complex aging studies. The research project aims to bridge this gap by developing new statistical machine learning methods for learning graphical models from heterogeneous and non-Gaussian observations and inferring changes in graphical models in different subpopulations. In particular, the project will (i) develop a flexible framework for estimation of multiple graphical models from heterogeneous populations with complex structures, (ii) develop an inference framework for detecting differential connectivity in biological networks, and (iii) provide a general framework for estimation of non-Gaussian graphical models, with changes over time or experimental conditions. Together, these tools provide a comprehensive framework for differential network analysis and will advance the current state of statistical machine learning methods for the analysis of high-dimensional biological networks.
