Statistical Methods for Molecular Evolution

Statistical methods for inference about evolution from aligned molecular sequence data will be the focus of this research. In addition, working with colleagues in biology and biochemistry, methods will be applied to better understand the evolution of early single-celled organisms and how processes of selection work in pathogens. While statistical theory and methods will be developed with evolutionary inference questions in mind, the statistical methods and theory will be more broadly applicable to other areas of science. Four project areas will be targeted: 1. Tree Testing: Understanding what the evolutionary relationships are between organisms is an important step in understanding their biology. We will derive new statistical methods that yield confidence sets of evolutionary trees that are likely to include the true relationship and provide information about where uncertainty about evolutionary relationships lies. 2. Positive Selection: This occurs when organisms adapt to changing environmental conditions. Detecting positive selection is of importance, for instance, in understanding how pathogens that infect humans can evolve resistance. Current methodology sometimes gives biologically unreasonable estimates of the strength of selection and can give unreliable testing results. We are developing methods that are less susceptible to the sparse-data issues that cause such problems. Another area of interest is developing models that jointly models changes in the functions that an organism's genes performs and changes in DNA. Such models will give a better understanding of which locations in a gene are important for particular biological functions. 3. Protein Evolution Models: Evolution of the proteins that perform the functions of organisms is a complex process. Understanding such evolutionary processes is crucial to inference about the relationships between organisms and of interest in itself. We will develop more realistic models that accommodate the frequently observed phenomenon that evolutionary processes vary over proteins and positions within proteins. 4. Model Selection: An important task in modeling evolution is to select classes of models that adjust for important processes but without becoming so complex that data is insufficient for their estimation. We will develop methods that penalize excess complexity and test model performance by considering key performance measures on validation data.