Optimization identification of superdiffusion processes in biology: an algorithm for processing observational data and a self-similar solution of the kinetic equation
This work is an attempt to transfer to biology the methods developed in physics for formulating and solving the kinetic equations in which the kernel of the integral operator in spatial coordinates is slowly decreasing with increasing distance and belongs to the class of Levy distributions. An algorithm is proposed for the reconstruction of the step-length probability density function (PDF) on a moderate number of trajectories of biological objects (migrants) and for the derivation of the Green's function of the corresponding integro-differential kinetic equation for the density of migrants in the entire space-time range, including the construction of an approximate self-similar solution. A wide class of time-dependent superdiffusion processes with a model power-law step-length PDF is considered, which corresponds to "Levy walks with rests" for given values of the migrant's constant velocity and the average time T of the migrant's stay between runs. The algorithm is tested within the framework of a synthetic diagnostics, consisting in the generation of artificial experimental data for trajectories of migrants and the subsequent reconstruction of the parameters of the step-length PDF and T. For different volumes of synthetic data, to obtain a general idea of the distributions under study (non-parametric case) and to evaluate the accuracy of recovering the parameters of the PDF (in the case of a parametric representation), the method of balanced identification is used. The approximate self-similar solution for the parameters of step-length PDF and T is shown to provide reasonable accuracy of the space-time evolution of migrant's density.