Information-Theoretic Framework for Allosteric Communication Pathway Prediction From Protein Structure.

Allosteric communication is fundamental to the function of proteins, but identifying the residue-level pathways through which signals travel and their directionality is still a major challenge. Here we present a multi-metric information-theoretic framework that identifies candidate allosteric communication pathways from a single protein crystal structure in seconds. The method combines Transfer Entropy (TE) from the Gaussian Network Model to assign a model-based directional asymmetry to residue-residue coupling, Mutual Information (MI) to quantify coupling strength, and Conditional Mutual Information (CMI) to detect relay residues that mediate indirect couplings. At the network level, Burton-Pemantle ensemble random spanning trees provide robustness-weighted betweenness centrality, differentiating persistent communication bottlenecks from occasional hubs. Kullback-Leibler (KL) divergence of the magnitude of the net transfer entropy distributions quantifies network rewiring and identifies allosteric switches. We validate the framework on four experimentally characterized systems: PDZ3 of PSD-95, p38α MAP kinase, KRAS WT versus the oncogenic G12D mutant, and kinesin-5 in ATP versus ADP bound states. Quantitative validation shows that the workflow captures previously reported hotspots and exceeds MI-only and contact graph shortest-path baselines on the all-paths overlap metric, while performing at least as well as topology-matched random controls. Computations take from ~14 s for a 110-residue domain to ~7 min for a 350-residue domain. on a standard laptop. Thus, the method can be applied systematically to large families of proteins.