FIGURE 4:

Our method’s accuracy increases as more prior terms are specified explicitly. We used low-noise experimental data from the Peterman and Wuite groups to test the effects of our prior. (A) No blinking or overlapping events are considered, and we integrate over all unknown means and variance to obtain the marginal posterior. Although some steps are found accurately, others are not, and both the double-step and the reactivation events are predictably missed. (B) Here we use the marginal posterior given in Eq. 7, which, as expected, shows event stacking and overestimation of the number of active fluorophores for the reasons described in Illustrative example: a model selection criterion assuming no event blinking or overlap. (C) Here we add the explicit form of the priors for P(K) and P(arr|m, K). The overestimation of fluorophore numbers problem is resolved, although the algorithm still gets the event stacking wrong, causing a step to be missed. (D) Our full posterior, Eq. 11. Results are excellent compared with A–C.