Statistical Models of Biopolymer Sequence and Folding

Duke University, Durham NC

Investigators

Abstract

Proposal ID: 0204690 PI: Scott Schmidler Title: Statistical models of biopolymer sequence and folding Abstract: This research involves development of new probabilistic and Bayesian statistical models for the analysis of biopolymer sequences. Emphasis is on predictive modeling of proteins and RNA. Models for sequences of random variables with complex short- and long-range interaction structure are developed and explored. Particular focus is placed unifying statistical models estimated from data with statistical mechanical models of polymer folding estimated via experimental parameter measurement. Targeted applications include protein structure prediction, protein folding kinetics, and protein-RNA binding. Statistical methodology development focuses on connecting statistical models for sequence analysis and change-point problems, including graphical Markov models and random fields, to statistical mechanical models of polymer folding, especially on biopolymers (proteins and RNA), to develop predictive theories. An additional core component of this research program concerns development of computational methodology for probabilistic inference in these models, including novel Markov chain Monte Carlo (MCMC) algorithms for multi-modal distributions and rough energy landscapes. Modern research in the molecular biosciences and biomedicine relies increasingly on both computational modeling and analysis of large collections of experimental data. This research concerns development of novel and unified methods for combining these areas. This work leverages physical models to develop improved methods for statistical data analysis, and uses statistical methodology for improving predictive accuracy of physical models. The focus is on analysis of protein and nucleic acid sequences and structures being generated by high-throughput whole-genome analyses. These advances will provide important new statistical methodology for computational biology, as well as provide domain scientists with improved tools for data analysis and predictive modeling.