Friday, December 1, 2017
11 AM - 12 PM
Machine learning is becoming the driving force for efficient and robust discoveries in biomedical research, given the vast amount of functional genomics datasets generated by high-throughput techniques. These heterogeneous panels of genome-wide 'omic' data provide the unique opportunity to systematically annotate the functional roles of specific genomic locations in diverse cellular contexts, differentiation stages and environmental conditions, which is a critical step to decode human disease mechanisms. In the meantime, sophisticated statistical models and machine learning algorithms are needed to address the 'big-data' challenges, such as pervasive correlations across high-dimensional data, non-linear associations, missing values, weak statistical power, biased samples and low signal-to-noise ratios.
In this talk, I will introduce multiple projects in my lab which decipher context-dependent gene regulation networks using hierarchical graphical models. I will first discuss a new iterative learning algorithm to improve the prediction of different kinds of regulatory elements along the whole human genome, especially in ambiguous repetitive genomic regions. The expanded map of regulatory elements serves as a foundation for network construction. I will then introduce a new probabilistic model and joint MCMC inference algorithm for 3D long-range enhancer-gene network predictions in diverse cellular contexts. We further leverage the predicted networks to discover interrupted biological pathways in 14 complex human diseases, including colorectal and breast cancer, which substantially improves our systems-level interpretation of the underlying disease mechanisms. Finally, I will present a new mixture model to delineate causal regulatory genetic variants in tumor development and the latent combinatorial regulatory grammar. The developed computational algorithms and the large-scale network predictions will provide better insights on gene regulation, 3D chromatin architecture and disease mechanisms, leading to novel genomics-based diagnostics and therapeutics.
Dr. Jianrong Wang is an Assistant Professor in the Department of Computational Mathematics, Science and Engineering (CMSE). His research is focused on developing statistical models and machine learning algorithms to infer large-scale context-dependent gene regulatory networks and leveraging the network predictions for human disease analysis. Major research topics in Dr. Wang's group include inference of long-range enhancer-gene networks and associated 3D chromatin structure, delineation of hierarchical regulatory grammar of gene expression, genome-wide regulatory element prediction based on high-dimensional epigenetic features, efficient signal processing for high-throughput biological dataset, and functional annotation of non-coding genetic variants in diverse human diseases.
Dr. Yanni Sun