Resolving FRET signal degeneracy and population heterogeneity via Bayesian nonparametrics

Recent experiments have revealed that the dynamics of individual biomolecules can be highly heterogeneous, with different molecules visiting different numbers of states and forming distinct subpopulations. This challenges analysis frameworks implicitly assuming an “average molecule” with fixed states and kinetics. While single molecule Förster resonance energy transfer (FRET) is widely used to probe such dynamics at the nanometer scale, existing analysis tools struggle to capture the molecule-to-molecule variability or distinguish states that differ primarily in their kinetics, rather than FRET signal, as FRET projects many 3D molecular configurations onto similar 1D signals. Accurate inference of molecular states from FRET data is particularly challenging for methods that require users to predefine the number of states or noise characteristics. Here we demonstrate a new framework, Bayesian nonparametric FRET for binned data (BNP-FRET-Bin), that minimizes user-tuned inference parameters such that assumptions are limited to those of hidden Markov models. Furthermore, BNP-FRET-Bin accurately incorporates many known noise sources in each FRET detection channel, thereby enabling the identification of distinct configurations from 1D traces via their kinetic signatures even when their FRET efficiencies overlap. Using simulated and experimental data, we demonstrate that BNP-FRET-Bin removes the logistical barrier of predetermining states for each FRET trace and permits high-throughput, simultaneous analysis of large numbers of biologically heterogeneous traces. Furthermore, working in the Bayesian paradigm, BNP-FRET-Bin naturally provides uncertainty estimates for all model parameters, including the number of states, kinetic rates, and their associated FRET efficiencies.