Vibrational entropy estimation can improve binding affinity prediction for non‐obligatory protein complexes. Issue 4 (21st January 2018)

• Record Type: Journal Article
• Authors: Škrbić, Tatjana; Zamuner, Stefano; Hong, Rolando; Seno, Flavio; Laio, Alessandro; Trovato, Antonio
• Is Part Of: Proteins. Volume 86:Issue 4(2018)
• Page Start: 393
• Page End: 404
• Publication Date: 2018-01-21
• Subjects: elastic network models -- molecular dynamics -- protein–protein interaction -- scoring functions -- statistical potentials; Proteins -- Periodicals; Proteins -- Periodicals; 572.6
• DOI: 10.1002/prot.25454
• Languages: English
• ISSNs: 0887-3585

Abstract

Abstract: Predicting the binding affinity between protein monomers is of paramount importance for the understanding of thermodynamical and structural factors that guide the formation of a complex. Several numerical techniques have been developed for the calculation of binding affinities with different levels of accuracy. Approaches such as thermodynamic integration and Molecular Mechanics/Poisson‐Boltzmann Surface Area (MM/PBSA) methodologies which account for well defined physical interactions offer good accuracy but are computationally demanding. Methods based on the statistical analysis of experimental structures are much cheaper but good performances have only been obtained throughout consensus energy functions based on many different molecular descriptors. In this study we investigate the importance of the contribution to the binding free energy of the entropic term due to the fluctuations around the equilibrium structures. This term, which we estimated employing an elastic network model, is usually neglected in most statistical approaches. Our method crucially relies on a novel calibration procedure of the elastic network force constant. The residue mobility profile is fitted to the one obtained through a short all‐atom molecular dynamics simulation on a subset of residues only. Our results show how the proper consideration of vibrational entropic contributions can improve the quality of the prediction on a set of non‐obligatory protein complexes whose binding affinity is known.