Large-scale conformational changes of protein molecules are ubiquitous in the biophysical and biochemical processes that sustain the life of a cell. Typical examples include cell signaling, intracellular transportation, transcription, and translation. Molecular level understanding of how protein molecules conduct conformational changes in responding to different environmental variables is thus critical in many areas of biotechnology and pharmaceutical industry. Although the advancement of atomistic empirical potential and configuration sampling methodologies have made significant contributions to the understanding of biomolecules, severe limitations still constrain the studies of large-scale protein conformational changes at the atomistic level. As a result, numerous reduced representations, or coarse-grained (CG) models, have been developed to study the biophysical properties of protein molecules. The Elastic Network Model (ENM) is one example and has been shown to be able to describe the low-frequency vibration modes of biomolecules and to predict the distribution of thermal fluctuations (X-ray B-factors) of a protein molecule. In this work, the ENM is extended to model the structural transition of a protein molecule by applying the empirical valence bond theory (EVB). EVB has been applied to construct the complicated potential energy surfaces (PES's) of enzymatic reactions and proton transfer in molecular liquids from a small number of high-level electronic structure calculations. By exploiting the capability of ENM in capturing the low-frequency motions, combining ENM and EVB thus provides a potential route in modeling the large-scale structural transition of a protein molecule. The numerical difficulties that result in non-physical cusp shapes of barriers in earlier attempts will first be described. A solution to overcome these difficulties will be proposed such that smooth PES's can be constructed with parameters of physical significance. The results of applying this ENM-EVB approach to describe the structural transitions of protein actin and adenylate kinase will then be presented. The potential of coupling this CG methodology with atomics level simulations will also be discussed.