Bayesian Methods for Protein Fibrillization: Model Integration and Network Dynamics

University Of California-Irvine, Irvine CA

Investigators

Abstract

This project centers on the development of principled statistical methods for understanding the formation and growth of amyloid fibrils, protein aggregates with broad functional and disease-related biological relevance. Protein fibrillization is a basic biophysical phenomenon that underlies problems of immense social concern. These include diseases such as Alzheimer's, cataract, and type II diabetes that are increasingly prevalent in our aging population, as well as economic costs and food security concerns related to prion disease in cattle and other non-human animals. The proposed research has the potential to inform the search for solutions to these serious societal problems, resulting in both economic savings and improvements in individual lives. By developing new predictive and data analytic techniques and validating them with novel experimental data, the project will advance our understanding of the factors that enhance or inhibit protein fibrillization while also producing statistical innovations that can be potentially applied to other problem domains. This project will also provide a unique interdisciplinary training program for graduate and undergraduate students, incorporating novel statistical methods, programming, and experimental techniques. The research project combines modeling techniques from the mathematical social sciences with theoretical and experimental methods from biophysical chemistry, enabling us to approach biological problems in novel ways. The technical innovations of this project are focused on two areas. First, it will develop new approaches to Bayesian model integration in which integrated predictions will be obtained from multiple, potentially non-statistical models in cases with little or no test data. Second, the project will develop novel model families for fibrillization kinetics, extending methods originally developed for social networks to capture interactions between individual proteins in solution over time scales of hours to days. The modeling work will be validated by combination of existing experimental data and by biophysical data collected by the research team. The research will result in new Bayesian techniques for predicting phenomena related to protein aggregation (especially in a high-throughput setting), and for modeling the kinetics of fibrillization process itself. The proposed research will also lead to novel methods for Bayesian integration of predictive models for phenomena with complex dependence, new Bayesian inference, model selection, and simulation techniques for large-scale dynamic network models, and a body of biologically relevant empirical data on protein fibrillization.