Robustness of ExTrack to biases due to wrong model assumptions. (a) Predictions of d₁, k_u, and k_b in case of two-state parameter fits to two-state simulations with one immobile state and one diffusive state. Position at each time point show variable localization errors σ. Peak-wise localization errors were specified to the model. σ followed a chi-square distributions re-scaled so the mean localization error equals 0.02 µm (for sample distributions, see inset of Fig. 2 b). Simulations with d₁ = 0.1 µm and k = k_u = k_b = 0.1Δt⁻¹. 10 replicates per condition. ExTrack settings: window length = 7, no sub-steps. (b) We considered tracks from simulated particles with one immobile state (⁠$d0*$ = 0) and five diffusive states with similar diffusion lengths of values 0.04, 0.06, 0.08, 0.1, and 0.12 µm (corresponding to $d*$ from 2 to 6), transition rates were set to randomly picked values (See Materials and methods, Computation simulation of tracks for more details on the transition rates values). This model results in indistinguishable diffusive tracks. Left: Distribution of displacements (for each dimension) of the five diffusive states. Right: Bar plots of true and estimated parameters obtained from fitting to a three-state model followed by aggregation of the diffusive states and computation of the resulting parameters. Here, the fractions are the global fractions computed from rates. See Table S2 for other results. (c) Heatmaps of relative errors on d₁, k_u, and k_b with variable diffusion coefficient following the same protocol than in Fig. 2 b for vbSPT and anaDDA.