We have developed different computational methods for refining, modeling and simulating protein structures in multi-scale resolution. Combining with experimental information from different structural biological method such as Fiber Diffraction, cryo-Electron Microscopy (cryo-EM), Small Angle X-ray Scattering (SAXS) and X-ray Crystallography, we show that our methods can be employed to improve the study of protein structures. In detail, the atomic structure of actin filament was refined against fiber diffraction data by our long-range normal mode based refinement protocol, which for the first time demonstrate that, for any fiber diffraction data, a substantial amount of refinement error is due to the deformations of the filaments. A geometry-based loop motif filter was then constructed followed up with an energetic ensemble optimization, in order to detect higher resolution structural information from intermediate-resolution cryo-EM experimental data. The results also imply that, among all the possible topology candidates for a given skeleton, evolution has selected the native topology as the one that can accommodate the largest structural variations, not the one rigidly trapped in a deep, but narrow, conformational energy well. We further introduced a new Monte-Carlo simulation technique, which combines elastic network model and a Hamiltonian at a different scale to study the protein folding problem assisted by the small angle X-ray scattering (SAXS) profiles. It was shown that our approach was effective for deriving the topology of small, globular helical proteins or protein domains. Finally, a new knowledge-based potential that only requires the Ca positions as input was built, which is expected to adds a new tool for protein structural modeling.