Figure 5
Parameter scatterplots for stochastic and deterministic Psp models. Inferred parameter distributions. Shown are the two-dimensional projections of the 10-dimensional intermediate and posterior parameter distributions, i.e. the output of the ABC SMC algorithm consisting of all accepted particles (i.e. parameter combinations). Circles correspond to accepted particles (k1, ..., k10), which result in a good fit to the data (see Figure 6). Eight ABC SMC populations were run, and particles from each population are coloured by a different colour. The particles of the last population are coloured in yellow - this population of particles approximates posterior parameter distribution, and its particles are parameter combinations that give the best fit of the model to the data (in a Bayesian sense). The parameter determining the damaged membrane was set to a = 60. (a) Parameters inferred in a stochastic frameworks. (b) Parameters inferred in a deterministic framework. The parameters in deterministic framework were sampled from the following priors: k1, k3, k4, k8, k9 ~ U (0, 1), k2 ~ U(0, 100), k5 ~ U(0, 0.05), k6, k7 ~ U(0, 0.01), k10 ~ U(0, 5). In the stochastic framework, corresponding priors were calculated as explained in section. Tolerance levels used in ABC SMC algorithm: ε1 = (100, 13.0, 100.0, 100.0, 1.5), ε2 = (80, 10.0, 100.0, 100.0, 1.3), ε3 = (60, 8.0, 70.0, 70.0, 1.2), ε4 = (50, 7.0, 60.0, 60.0, 1.1), ε5 = (40, 6.0, 50.0, 50.0, 1.0), ε6 = (30, 5.0, 40.0, 40.0, 0.9), ε7 = (20, 4.0, 30.0, 30.0, 0.8), ε8 = (10, 3.0, 20.0, 20.0, 0.7). These tolerance levels together with the distance function (d1, ..., d5) defined in the text, determine which proposed particles will be accepted.