Fig. 4: Resolving kinetic heterogeneity: states with indistinguishable FRET efficiencies but different kinetics.

a An illustration of the simulated GT model with states (circles) connected by forward and reverse rate constants. States 1 and 4 as well as states 2 and 3 have indistinguishable FRET efficiencies, causing kinetic heterogeneity. b Two simulated FRET traces offset in time with donor and acceptor fluorescence intensity (green, red) and FRET efficiency (FRET, black) are shown, representative for the dataset used in (c, d): n(traces) = 250, n(datapoints) = 56,794, sampling rate = 5 Hz, time per datapoint = 200 ms. c FRET histogram with inferred FRET efficiencies overlaid (right axis: legend as in Table 1 and in all Figures). Sample size as in (b). The error bars represent standard deviations. d Comparison of cumulative dwell time distributions derived from the kinetic models with two FRET states (detailed in Methods). The GT histogram is shown as a bold black line. Insets show zoomed-in views of the data indicated by the squares. e Quantitative comparison of the four most accurately inferred kinetic models: the GT values are represented as red and blue vertical lines. Sample size as in (b). Uncertainty measures (CI, SD) as listed in Table 1. The intrinsic uncertainty of the dataset is shown as dark gray (1σ) and light gray (2σ) intervals. Beyond the six displayed rate constants, these additional rate constants were inferred: for Hidden Markury k₃₁ = 0.045 and k₃₄ = 0.003, for SMACKS k₁₃ = 0.0001, k₃₁ = 0.0055, k₃₄ = 0.0034, for MASH-FRET (prob.) k₃₁ = 0.033. All inferred values of all models are reported in the Supplementary Tables 1 and in the Supplementary Datafiles. ‡ denotes results that were submitted after the GT was known. No results were reported by Edge finding. Participants were informed that kinetic heterogeneity may be involved, but not in which configuration. Source data are provided as a Source Data file for panels (c, d, and e).