Prediction of H-Bond Rotations from Protein H-Bond Topology

TL;DR

This paper investigates the relationship between H-bond geometry and topology in proteins, proposing two prediction methods that accurately estimate H-bond rotations from subgraph features, revealing insights into protein folding.

Contribution

It introduces two novel approaches to predict H-bond rotations from protein topology, enhancing understanding of protein structure beyond traditional methods.

Findings

88.14% of predictions are within the true rotation in SO(3)

Prediction methods occupy only 1% of the rotation space volume

Topology-based predictions outperform random baselines

Abstract

H-bonds are known to play an important role in the folding of proteins into three-dimensional structures, which in turn determine their diverse functions. The conformations around H-bonds are important, in that they can be non-local along the backbone and are therefore not captured by the methods such as Ramachandran plots. We study the relationship between the geometry of H-bonds in proteins, expressed as a spatial rotation between the two bonded peptide units, and their topology, expressed as a subgraph of the protein fatgraph. We describe two experiments to predict H-bond rotations from their corresponding subgraphs. The first method is based on sequence alignment between sequences of the signed lengths of H-bonds measured along the backbone. The second method is based on finding an exact match between the descriptions of subgraphs around H-bonds. We find that 88.14% of the predictions lie inside the ball, centred around the true rotation, occupying just 1% of the volume of the rotation space SO(3).