Efficient Sampling in Stochastic Biological Models

[摘要] Even when the underlying dynamics are known, studying the emergent behavior of stochastic biological systems in silico can be computationally intractable, due to the difficulty of comprehensively sampling these models.This thesis presents the study of two techniques for efficiently sampling models of complex biological systems.First, the weighted ensemble enhanced sampling technique is adapted for use in sampling chemical kinetics simulations, as well as spatially resolved stochastic reaction-diffusion kinetics.The technique is shown to scale to large, cell-scale simulations, and to accelerate the sampling of observables by orders of magnitude in some cases.Second, I study the free energy estimates of peptides and proteins using Markov random fields.These graphical models are constructed from physics-based forcefields, uniformly sampled at different densities in dihedral angle space, and free energy estimates are computed using loopy belief propagation.The effect of sample density on the free energy estimates provided by loopy belief propagation is assessed, and it is found that in most cases a modest increase in sample density leads to significant improvement in convergence.Additionally, the approximate free energies from loopy belief propagation are compared to statistically exact computations and are confirmed to be both accurate and orders of magnitude faster than traditional methods in the models assessed.

[发布机构] the University of Pittsburgh