Belief Propagation Reveals Allosteric Mechanisms in Proteins

The physical mechanisms involved in allosteric regulation remain unclear. We present a novel and efficient method for investigating the propagation of regulatory signals in protein structures. Our approach utilizes undirected graphical models to efficiently encode the Boltzmann distribution over geometric configurations. Belief Propagation is then invoked to efficiently compute: (a) free energies and (b) allosteric couplings between distal residues. We present results from two kinds of experiments. First, we show that our method accurately predicts changes in free energy upon activation and/or mutation. Specifically, our method achieves a high correlation with experimentally determined ΔΔGs (R² = 0.90 for core residues). Significantly, our method is capable of identifying those residues experiencing the largest relative changes in enthalpy and/or entropy. Second, we use our method to study the allosteric behavior of cyclophilin A in enzyme catalysis. Our analysis reveals the allosteric coupling between residues separated by as much as 20 angstroms from the active site. These results correspond well with experimental measurements. Our method requires a few minutes per protein, making it suitable for large-scale studies. Taken together, these results suggest that our method provides an effective means for investigating allosteric regulation at the proteome scale.