Research
Consider ice, water, and steam. They each have very different macroscopic properties but are composed of the same molecules. The study of statistical physics has shown that the way the individual water molecules interact with each other leads to these dramatic changes, or “phase transitions,” in macroscopic behavior. Using multi-‘omics data and mathematical analysis, we have found that the way genes interact with each other in large, complex cellular networks can drive changes in the overall phenotype of an organism. Building on this work, we are developing a set of computational methods to leverage biological “Big Data” and identify features of regulatory networks that control the interplay between disease, genetic variation, tissue identity, and drug response.
Detecting changes in network modular structure
Genes that work together to perform biological functions tend to form network modules. But biological networks are dense, noisy, and hierarchical, making it difficult to know which modules are active in a given context. We have created a method called ALPACA that can tease apart disease modules within large transcriptional networks. We are using ALPACA to find predictive patterns in genetic variation, pathways underlying viral tumorigenesis, and tissue-specific regulatory circuits.
Finding drivers of disease
An important question in biological networks is: how should we perturb the network to alter the phenotype and reverse disease? We have found that some regulators, which have many protein binding partners and regulate many genes, act as bottlenecks in the information highways of the cell and are more likely to be drivers of phenotype. Ultimately, we need to build predictive models that incorporate changes in node and edge activity and infer the interventions that can direct the cell from a disease state to a healthy one.
Modeling drug response in cancer
Predicting how different individuals respond to drugs is essential for precision medicine. We are building Bayesian network models that incorporate mutation status and tissue specificity to predict drug response in cancer cell lines.
new algorithms for biological network analysis
Traditionally, network science has been developed in the context of macroscale networks and abstract graphs. In biology, we need a new theory of networks that incorporates from the ground up the dynamic, fluctuating interactions between molecular components of the cell. We are working towards such a theory using concepts from statistical physics to characterize ensembles of networks and their phase transitions.