AlloPath: A method for identifying protein allosteric pathway based on transfer entropy and hidden Markov model

Allostery is one of the most direct and efficient mechanisms for regulating protein functions. Allosteric pathway is the core of the mechanism, which serves as a communication route to transmit perturbations from allosteric site to active site. Developing theoretical methods to identify protein allosteric pathway has been a challenging but interesting task. Here, we propose an effective approach AlloPath to identify allosteric pathway, where the transfer entropy (calculated based on Gaussian Network Model) with the time-delayed correlation considered is utilized to quantify the strength of information transfer between neighboring residues, and the Hidden Markov Model and dynamic programming method Viterbi are employed to produce and identify an optimal information transfer pathway in allostery. AlloPath is tested on three well-studied classic allosteric proteins: hPTP1E PDZ2, Caspase-1 and CheY, and the prediction results align closely with the existing experimental and theoretical data. For hPTP1E PDZ2, the three directional allosteric pathways linking the peptide-binding pocket to three distal surfaces (DS1, DS2, DS3) are accurately predicted. For Caspase-1, a long-distance allosteric pathway (∼30 Å) extending from active site 1 to active site 2 via allosteric site has been successfully identified. For CheY, the allosteric pathway mediating the phosphorylation signal transmission from D57 site to the N16-FliM binding pocket is accurately captured. Additionally, the allosteric pathways predicted by AlloPath are of high reversibility and robustness. AlloPath relies solely on the static topological structure of protein, eliminating the need for costly computation, which makes it a powerful tool for identifying protein allosteric pathways.