Optimal inference of molecular interactions in live FRET imaging

Intensity-based live-cell fluorescence resonance energy transfer (FRET) imaging converts otherwise unobservable molecular interactions inside cells into fluorescence time-series signals. However, inferring the degree of molecular interactions from these observables is challenging, due to experimental complications such as spectral crosstalk, photobleaching, and measurement noise. Conventional methods solve this inverse problem through algebraic manipulations of the observables, but such manipulations inevitably accumulate measurement noise, limiting the scope of FRET analysis. Here, we introduce a Bayesian inference framework, B-FRET, which estimates molecular interactions from FRET data in a statistically optimal manner. B-FRET requires no additional measurements beyond those routinely conducted in standard 3-cube FRET imaging methods, and yet, by using the information contained in the data more efficiently, dramatically improves the signal-to-noise ratio (SNR). We validate B-FRET using simulated data, and then apply it to FRET data measured from single bacterial cells, a system with notoriously low SNR, to reveal signaling dynamics that are otherwise hidden in noise.