Bayesian Nonparametric Models for Freely Diffusing SmFRET Data

• Activity Number: 91 - High Dimensional Data, Causal Inference, Biostats Education, and More
• Type: Contributed
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: ENAR
• Author(s): Roberto Rivera* and Jared Hidalgo and Axel Cortes Cubero and Roberto Reyes
• Companies: University of Puerto Rico at Mayaguez and University of Puerto Rico - Mayaguez and University of Puerto Rico at Mayaguez and University of Puerto Rico at Mayaguez
• Keywords: Bayesian nonparametric; smFRET; Dirichlet process; Markovian model; freely diffusing; disentangled-sticky model

Abstract

There is great interest in developing a fundamental understanding of conformational dynamics of enzymes. The details of these structural evolutions is mostly unexplored due to the lack of experimental characterization tools with sufficient spatial and temporal resolutions and the inherent complexity of the systems. Single molecule fluorescence resonance energy transfer (smFRET) experiments can in principle directly probe the conformational states of a biomolecule at each moment in time. However, smFRET experiments are fraught with shot-noise, background photons, and cross-talk making the signal to noise ratio very low. Statistical analysis is needed to quantify the number of FRET states, and the transition rates between them. Bayesian nonparametric models guard against gross misspecification of the data-generating mechanism and instead of providing a fixed number of states, probability models with an infinite dimensional space of states are deployed. We also explore one Bayesian nonparametric model that accounts for the artefacts influencing smFRET experiments. The models as well as the conventional hidden Markov model are applied to simulated and experimental data.