Directional information flow as a tool for analyzing protein allostery

Dynamical network analysis is an invaluable tool for quantifying protein allostery. Existing workflows that apply dynamical network models to molecular dynamics simulations use time-symmetric metrics for identifying correlated motions between residue pairs. However, these methods do not capture the directionality of information flow from which causal relationships can be determined. In recent years, the transfer entropy measure of information flow has been applied to study causal relationships in protein dynamics. We used transfer entropy to develop a workflow for generating causal protein networks based on residue fluctuations and applied it to study allosteric communication in the SARS-CoV-2 main protease, a protein that binds allosteric ligands at various sites. We identified directional information flow between residue contacts emerging at various time lags and determined whether residues generate or receive information from their neighbors. Lastly, we used our transfer entropy network to quantify the directionality of optimal paths, allowing us to determine causal relationships between important binding sites of substrate ligands or drug molecules.