Both methods converge to the global solution and aim to maximize the convergence speed.
Mathematical conditions that maximize the convergence rate are derived and optimal schemes identified.
W is the ratio of the maximum ELR set up to maximize the convergence.
Consequently, we also restrict the role of the weighting factor to maximize the convergence speed of the LPIC detector and use only the stage index p as a regularization parameter.
Nevertheless, in practice, the role of the weighting factor is confined to maximize the convergence speed and not to control the amount of noise enhancement introduced into the solution.
In view of (7), the value of γ that maximizes the convergence rate is (13) γ = 2 ‖ Υ ‖ + ‖ Υ − 1 ‖ − 1, because this is the value that minimizes ‖ I − γ Υ ‖.
Upon convergence of the EM algorithm, the estimation procedure is switched to another iterative numerical optimization method in order to pinpoint the actual maximum by maximizing the log likelihood function, since convergence of the EM algorithm may be rather slow [ 26].
The control law is then optimized by maximizing the iLF-based lower bound of the convergence rate.
This process is repeated until convergence (i.e. the sites maximizing the PWM score are fixed in all sequences) or for at most a default number of 10 iterations, which we observed to often be sufficient for the convergence of significant seeded motifs.
We have also specified particular combinations of identifiable parameters, and these should be used in model fitting to avoid obvious numerical problems, of lack of convergence and absence of a unique set of parameters maximizing the likelihood.
