A general method for predicting amino acid residues experiencing hydrogen exchange

Abstract

Information on protein hydrogen exchange can help delineate key regions involved in protein-protein interactions and provides important insight towards determining functional roles of genetic variants and their possible mechanisms in disease processes. Previous studies have shown that the degree of hydrogen exchange is affected by hydrogen bond formations, solvent accessibility, proximity to other residues, and experimental conditions. However, a general predictive method for identifying residues capable of hydrogen exchange transferable to a broad set of proteins is lacking. We have developed a machine learning method based on random forest that can predict whether a residue experiences hydrogen exchange. Using data from the Start2Fold database, which contains information on 13,306 residues (3,790 of which experience hydrogen exchange and 9,516 which do not exchange), our method achieves good performance. Specifically, we achieve an overall out-of-bag (OOB) error, an unbiased estimate of the test set error, of 20.3 percent. Using a randomly selected test data set consisting of 500 residues experiencing hydrogen exchange and 500 which do not, our method achieves an accuracy of 0.79, a recall of 0.74, a precision of 0.82, and an F1 score of 0.78.

Computational Modeling of Molecular Structures Guided by Hydrogen-Exchange Data

“…Using a random forest classifier, Wang et al propose a machine learning method that can predict whether a given residue will have undergone HDX, based on experimental conditions and protein structure . From the Start2Fold database, they obtain information on the HDX status of residues, as well as experimental temperature and pH.…”

“…Using a random forest classifier, Wang et al propose a machine learning method that can predict whether a given residue will have undergone HDX, based on experimental conditions and protein structure. 204 From the Start2Fold database, they obtain information on the HDX status of residues, as well as experimental temperature and pH. From the protein structure, they extract features related to secondary structure, H-bonding, packing density, solvent accessibility, and Coulombic interactions.…”

“…Some of the limitations of structure-based HDX prediction models lie with inconsistencies in the molecular features that govern protection from HDX. These inconsistencies originate from the various types of conformational change that promote exchange, including global unfolding, subglobal motions intuitive interpretation, but weak correlation with HDX data, and challenges in defining SASA granularity hydrogen bonding 58,112,113,115,117 widely recognized as the main factor driving HDX protection, but not enough on its own to fully characterize protection electrostatic calculations 118,132,133,140 can explain the protection of solvent-accessible residues not involved in H-bonds, but force-field dependent and computationally intensive phenomenological expression [144][145][146][147][148]150 combines hydrogen bonding and atom burial; one of the most popular HDX prediction models, but can show poor correlation with HDX data combination of molecular features 55,[162][163][164]167,174 can involve nonstructural features (e.g., bond order, catalyst concentration, or hydrophobicity), but often just increases model complexity fractional-population models 55,107,172,179,180,182 estimate the free energy of exchange by calculating the probabilities of protected vs unprotected states, but very computationally intensive knowledge-based predictions 197,201,202,204 derive HDX protection directly from protein sequence, but usually show very weak correlation with HDX data between local minima, and breathing motions within minima. 58 The paradigm referred to as the fractional population model was proposed to circumvent difficulties associated with defining structure-based HDX prediction models.…”

“…Attempts to derive HDX data directly from protein sequences have largely proved unsuccessful, ,, but these methods have been included in this article for completeness. On the other hand, the HDX database Start2Fold has been used to develop machine learning algorithms that can predict the likelihood of amino acids being involved in early folding events. , This could form the foundation of methods capable of predicting HDX data from the primary structure, although the challenge facing knowledge-based predictions are still considerable and possibly exceed those faced by protein folding algorithms. The Start2Fold database raises an additional and important point, however, regarding an overall deficiency of adequately curated information on experimental HDX data for proteins with known structures.…”