Fig. 1

Overview of the procedure for building a structure-based Markov random field (MRF) model and quantifying coevolutionary constraints. The MRF architecture is determined from the three-dimensional structure of a target protein. Each residue of the target protein is represented as a node, and each intramolecular contact is represented as an edge connecting the corresponding residue nodes. After the graphical topology is determined, parameterization is performed for the multiple sequence alignment in the same way as in conventional MRF methods. Then, pairwise and positional coevolution scores are calculated from the edge and node weights, respectively