Neural relational inference to learn long-range allosteric interactions in proteins from molecular dynamics simulations

Abstract

Protein allostery is a biological process facilitated by spatially long-range intra-protein communication, whereby ligand binding or amino acid change at a distant site affects the active site remotely. Molecular dynamics (MD) simulation provides a powerful computational approach to probe the allosteric effect. However, current MD simulations cannot reach the time scales of whole allosteric processes. The advent of deep learning made it possible to evaluate both spatially short and long-range communications for understanding allostery. For this purpose, we applied a neural relational inference model based on a graph neural network, which adopts an encoder-decoder architecture to simultaneously infer latent interactions for probing protein allosteric processes as dynamic networks of interacting residues. From the MD trajectories, this model successfully learned the long-range interactions and pathways that can mediate the allosteric communications between distant sites in the Pin1, SOD1, and MEK1 systems. Furthermore, the model can discover allostery-related interactions earlier in the MD simulation trajectories and predict relative free energy changes upon mutations more accurately than other methods.

Origin of Correlations between Local Conformational States of Consecutive Amino Acid Residues and Their Role in Shaping Protein Structures and in Allostery

“…As discussed in the section “Role of the Along-Chain Correlations in Shaping Protein Structures and in Relaying Conformational Changes”, the reduced free-energy cost of the concerted rotation about the backbone virtual bonds is likely to contribute to allosteric interactions. This is in agreement with the view of allosteric networks as composed of the shortest connections between the units that relay a conformational change. ,, In this regard, it seems to be possible to employ the mathematical formalism developed in this work to identify allosteric networks composed of noncovalently interacting elements (e.g., the side chains) or propagating through helical segments, which can be coarse-grained to nearly linear segments by applying the model of Dawid and co-workers, in which a turn of a helix is a unit . Research in these directions is planned in our laboratory.…”

“…The concept of allosteric interactions was originally restricted to indirect interactions between different chains of multichain proteins such as, e.g., hemoglobin; , however, it turned out to be an intrinsic property of all proteins . One of the present views of allostery is that the change of the distribution of conformational states at one site induces that at the other site. , The mechanism of how the change is relayed has been studied by molecular dynamics. , Recently Zhu and co-workers combined molecular dynamics with neural network analysis to find the connection networks. These studies were focused on finding the networks of interacting side chains.…”

“…The noncovalent side chain–side chain interactions connecting the side chains would then play a role of covalent backbone interactions. It has been suggested by other researchers that the interacting side chains should form the shortest (presumably a straight) path for the allosteric interactions to be effective.…”

Protein Function Analysis through Machine Learning

“…Deep neural networks have been applied to short MD trajectories using a specialized metadynamics routine, called neural relational inference MD. In this model, an allosteric signal is put through a VAE to interpolate conformations and determine the communication pathways within the protein [ 342 ]. The combination of extreme gradient boosting (XGBoost) and GCNNs for allosteric site prediction was found to predict conformational allosteric sites on static structures tested, without a need for heavy simulation computations after being trained on known allosteric proteins to learn topological connections that define the phenomenon [ 343 ].…”

Deciphering the QR Code of the CRISPR-Cas9 System: Synergy between Gln768 (Q) and Arg976 (R)

“…Herein, we seek to find whether these mutations also alter the allosteric communication between the catalytic domains. These mutations determine the sensitivity to mismatches along the sgRNA–DNA hybrid duplex and are probed herein by computational approaches like classical molecular dynamics (MD), Markov state modeling (MSM), , an enhanced sampling technique, , and machine (deep) learning algorithms. − We focus on the dynamics of the catalytic His840 (HNH) and Arg976 (RuvC) residues, along with the Gln768 dynamics belonging to the linker domain between HNH and RuvC for the Cas9 wild type and mutants. The Arg976/Gln768 dynamics are also determined in relation to the presence of Mg 2+ ions that are indispensable for the action of Cas proteins .…”