Machine learning models designed for protein engineering and design typically rely on sequence-based, structure-based, or integrated representations that combine both sequence and structural information of proteins. These representations are used upon the domain knowledge: namely, that the structure (or the sequence) of a protein is partly responsible for its function. Despite their strength in capturing the evolutionary process of proteins, these representations often fall short of capturing the dynamic behavior of the protein structure. we propose a representation that incorporates knowledge of the protein's dynamic behavior obtained from molecular dynamics simulations. Our representation utilizes an unsupervised approach to observed time-series data generated from molecular dynamics simulations to encode both temporal and spatial dynamic behavior of a protein structure. Our dynamic-aware representation extracts essential relational interactions within the polymer chain revealing the interactions of sub-units of the protein which could be used to inform design strategies for protein engineering goals.