As start parameters, Brownian motion for microsatellite data was used, theta and M values were estimated from Fst calculations, Markov chain sampling: short chains 100, long chains 20.
For both models, we tested a wide range of start parameters for ω (0.01-5.0) and κ (0.5-5.0 0.5-5.0
All 24 markers, including paired interactions, physiological age, and the interaction between marker and physiological age, were used as start parameters.
All clusters get as start parameters for μ j,R, μ j,T the original retention time and IRM of peak location X j, respectively, for j=1,…, n.
Moreover, it is also found that the free flow self-organizes into grid-like structure when τ is large and the slow-to-start parameter 0<p<1.
Nevertheless, when the slow-to-start parameter p= 0, and traffic light parameter τ="3 or 5, it is found that phase separation phenomenon does not occur.
With carefully chosen starting parameter values, marginalized finite mixture models can be fitted using quasi-Newton optimization.
To ensure that the analyses had converged properly, we repeated each analysis three times from different starting parameter options and under different codon models.
For the best-fit model we re-ran the model four times with differing starting parameter values and found the results were unchanged.
Local analysis may simply test a particular fold-change in parameters (singly or multiply) from a single starting parameter set [26], [41], [42], whereas global analysis tests parameter sets that sample a defined region of parameter space [43].
"Global" analysis methods avoid this limitation by testing many starting parameter sets, to derive broader conclusions about the circuit of the model rather than the particular dynamics of one parameter set.
