In this work we propose a methodology for improving dynamical sampling in molecular simulations via temperature acceleration. The proposed approach combines the novel methods of Voter for temperature-accelerated dynamics with the multiple histogram reweighting method of Ferrenberg and Swendsen, its dynamical extension by Nieto-Draghi et al., and with hazard plot analysis, allowing optimal sampling with small computational cost over time scales inaccessible to classical molecular dynamics simulations by utilizing the "inherent structure" idea. The time evolution of the system is viewed as a succession of transitions between "basins" in its potential energy landscape, each basin containing a local minimum of the energy (inherent structure). Applying the proposed algorithm to a glass-forming material consisting of a mixture of spherical atoms interacting via Lennard-Jones potentials, we show that it is possible to perform an exhaustive search and evaluate rate constants for basin-to-basin transitions that cover several orders of magnitude on the time scale, far beyond the abilities of any competitive dynamical study, revealing an extreme ruggedness of the potential energy landscape in the vicinity of the glass transition temperature. By analyzing the network of inherent structures, we find that there are characteristic distances and rate constants related to the dynamical entrapment of the system in a neighborhood of basins (a metabasin), whereas evidence to support a random energy model is provided. The multidimensional configurational space displays a self-similar character depicted by a fractal dimension around 2.7 (+/-0.5) for the set of sampled inherent structures. Only transitions with small Euclidean measure can be considered as localized.