Towards full Quantum‐Mechanics‐based Protein–Ligand Binding Affinities. Issue 8 (6th March 2017)

• Record Type: Journal Article
• Authors: Ehrlich, Stephan; Göller, Andreas H.; Grimme, Stefan
• Is Part Of: Chemphyschem. Volume 18:Issue 8(2017)
• Page Start: 898
• Page End: 905
• Publication Date: 2017-03-06
• Subjects: ab initio calculations -- density functional calculations -- drug design -- noncovalent interactions -- proteins; Chemistry, Physical and theoretical -- Periodicals; 541.05
• DOI: 10.1002/cphc.201700082
• Languages: English
• ISSNs: 1439-4235

Abstract

Abstract: Computational methods play a key role in modern drug design in the pharmaceutical industry but are mostly based on force fields, which are limited in accuracy when describing non‐classical binding effects, proton transfer, or metal coordination. Here, we propose a general fully quantum mechanical (QM) scheme for the computation of protein–ligand affinities. It works on a single protein cutout (of about 1000 atoms) and evaluates all contributions (interaction energy, solvation, thermostatistical) to absolute binding free energy on the highest feasible QM level. The methodology is tested on two different protein targets: activated serine protease factor X (FXa) and tyrosine‐protein kinase 2 (TYK2). We demonstrate that the geometry of the model systems can be efficiently energy‐minimized by using general purpose graphics processing units, resulting in structures that are close to the co‐crystallized protein–ligand structures. Our best calculations at a hybrid DFT level (PBEh‐3c composite method) for the FXa ligand set result in an overall mean absolute deviation as low as 2.1 kcal mol −1 . Though very encouraging, an analysis of outliers indicates that the structure optimization level, conformational sampling, and solvation treatment require further improvement. Abstract : Absolute binding free energies for protein–ligand systems are calculated using a fully quantum‐mechanics‐based approach. The methodology works with large cutouts of protein binding pockets and is shown to yield promising results for two different protein targets. Although it is not yet fit for industrial application, the ab initio nature of the approach allows for clear identification of error sources. Potential improvements of the method are highlighted.