Fitting the model parameters

When chosing model fitting, clicking on Next will open the fitting window:

Number of states

To fit the model to the data, the first thing to do is to select a number of states. Then the initial estimates of all the individual parameters can be modified by clicking on the button Open Parameter Window. See Parameter Window for more informations on the parameters.

Frame time (in s)

Provide the time in between frames.

Window length

Provide the window length for which no approximations will be made. Outside the window, similar sequences of states will be merged according to their similarity. A higher window length increases the computation time.

Number of sub-steps:

Number of sub-steps where state-transitions can happen. Normally set to 1. Can be set to 2 or 3 if studying very transcient states (transition probabilities per step superior to 30%) but increasing this number has a stong impact on the fitting speed.

Threshold

Threshold for which similar sequences of states are merged. Increasing the threshold will speedup the method but decrease the quality of the fit.

Maximum number of sequences

Maximum number of sequences of states to consider. If that number is reached, ExTrack will increase the threshold to keep a number of sequences close to that number. Increasing that number improves the quality of the fit at the expense of speed.

Depth of field

Dept of field of the sample. This metric can be important to quantify the probability of tracks to leave the field of view and to therefor avoid the defocalization bias. This parameter assums wide field (or HALO) illumination that allows cytoplasmice tracks to leave from the top or bottom of the depth of field. If you are imaging in TIRF, correct the field of view by multiplying it by 0.7. If you are imaging particles that never leave the depth of field, put a high number (example: 100 * sqrt(2Dtime step)). If you are imaging membrane proteins, you can try to estimate the distance that a particle needs to cross to leave the field of view. If the depth of field is unknown, you can simply put an high value and ignore it. The best way to avoid the defocalization bias is to consider a maximum of track lengths (examples tracks of length 3 to 50).

Number of iterations

The fitting model may sometimes stop before the full convergence especially if the number of parameters is high. The algorithm prints the likelihood at each iteration. Pick a higher number of iterations if you see that the likelihood keeps improving at the last iteration.

Save path

Save path for the estimated model parameters. F0, F1, Fi correspond to the probability that a given track starts in the state i. This can be different from the actual fractions of particles in state i due to the effect of the defocalization bias. A better estimate of the actual fraction of each state is provided by the equilibrium_Fi. The latter is the fraction at equilibrium considering the transition rates of the model. The values pij correspond to the probability to transition from state i to state j per step. The corresponding transition rates per second can be computed by dividing pij by the time between subsequent frames.