Error on two-state model parameters for different methods. (a) Heat maps of mean relative error on extracted parameters from the same two-state simulations as in Fig. 1 for vbSPT and anaDDA. In the case of anaDDA, 20 replicates were used to assess its heat maps instead of 10 due to its higher variability. (b and c) Plots of estimated transition rates as a function of the actual rates for a subset of the simulations in a. (b) Results from UncertainSPT for different diffusion lengths. Gray curve: Ideal model. (c) Results from a modified version of ExTrack with a time-discretization approach (labeled Classical), which assumes transitions to occur at time points of molecule observations, and with our approach (labeled ExTrack), which assumes transitions to occur at the middle of each time step (see Materials and methods, Parameter fitting...). Tracks simulated with $d0*$ = 0, $d1*$ = 5. (d) Error, SD, and bias of parameters predicted by ExTrack, vbSPT, and anaDDA depending on the number of tracks in case of two-state tracks (five positions per track, k_u = k_b = 0.1 Δt, $d0*=0$⁠, and $d1*=5$⁠). The error (root mean square error) can be decomposed into bias (absolute value of the difference between the average estimate from all replicates and true parameter) and SD (std) of the estimated parameters. Error = $bias2+std2.$ Obtained from 100 replicates. Here, all estimated values are relative to their true value. ExTrack settings: number of sub-steps = 2, window length = 10.