A combined study of molecular dynamics simulations, information theory, molecular graph theory, and machine learning in protein structure and dynamics.

Using a notably large amount of data to investigate physical and chemical phenomena requires new statistical and computational approaches; besides, the cross-validations require well-established theoretical frameworks. This study aims to validate the statistical efficiency of alternative definitions for information-theoretic measures, such as transfer entropy, using the so-called [Formula: see text]-framework. The primary goal is to find measurements of high-order correlations that preserve information-theoretic properties of information transfer between the components of a dynamical system (such as a protein) due to local operations. Besides, this study aims to decode the information contained in the amino acid sequence, establishing a three-dimensional protein structure by comparing the amino acids’ physical-chemical properties with their ranked role in the protein interaction network topology using new graph-theoretic measures based on the constructed digraph models of [Formula: see text] information transfer within a heat flow kernel embedding framework. Moreover, this study aims to utilize Deep Graph Convolutional Neural Networks for classifying the role of each amino acid in a protein trained upon short equilibrium structure fluctuations at sub-nanosecond time scales. In particular, this study examines the effect of disulfide bridges on the three-dimensional structure of wild-type and mutated analogs of bovine pancreatic trypsin inhibitor protein.