PhD Conferral Eko A. Rifai
Aula, Vrije Universiteit Amsterdam
Applicability and performance of linear interaction energy based binding affinity calculation
Eko A. Rifai
Prof. dr. Nico P.E. Vermeulen, copromotor dr. Daan P. Geerke
Amsterdam Institute of Molecular and Life Sciences
Evaluating and predicting the affinity of protein binding of drug-like compounds is directly relevant to pharmaceutical sciences in general and to drug discovery and development in particular. The protein-binding affinity governs ligand complexation with target proteins, and being able to quantitatively understand and predict this property can greatly support lead finding and/or optimization in the drug discovery process. Hence, improved efficiency and accuracy of computer-aided protein-ligand affinity methods are needed to increase the success rate of drug discovery and design and to save time and resources.
Since the 1990s, end-point methods have been serving as alternative methods for protein-ligand binding free energy calculation over the previously established rigorous alchemical methods and approximated docking methods, by combining conformational sampling and solvent effects (typically in MD) with a relatively fast scoring approach. The linear interaction energy (LIE) method is one of them and considers van der Waals and electrostatic interaction energies between ligand and surroundings from simulations of protein-bound and free ligands in water. Curated experimental binding free energies are used to parameterize alpha and beta as coefficients for van der Waals and electrostatic energies, respectively, which in turn constitute an equation that can be used to predict binding free energy of ligands with unknown affinities towards their target protein. With fitting of alpha and beta parameters in LIE, it is possible to calculate absolute binding free energies, and it in turn enables the inclusion of multiple binding poses with a Boltzmann-like weighting scheme as a single calculation of binding affinity per ligand. This makes LIE advantageous for calculating binding free energies of flexible proteins that can bind ligands in multiple orientations. However, the need of experimental parameters and the focus on end points only can affect the applicability of LIE modeling.
This thesis describes the development, evaluation, comparison and applicability of linear interaction energy (LIE) models for efficient computation of protein-ligand binding affinities in terms of their free energy of binding. With this work, we demonstrated that applicability domain (AD) analysis can be an estimation of LIE model reliability (even in the context of blind affinity prediction). Moreover, the estimated confidence levels of predictions for query compounds were in line with the correspondence between their protein-ligand interaction profiles and the interaction profiles obtained for the training compounds. This illustrates that applicability domains can rather depend on protein-ligand interactions covered during MD instead of properties of the ligands alone. We also confirmed that depending on the system of interest, the central LIE equation for scoring may be further optimized and the selection of docking poses can be crucial. In addition, we showed that correlations between calculated and experimental binding affinities obtained using LIE can be improved by combining it with alchemical solvation free energy calculations of the unbound ligands.