On the estimation of ligand binding affinities

Abstract: An accurate estimation of the binding free energy of a chemical reaction would be indispensable in numerous fields. For instance in drug design, a lot a time and resources could be saved by calculating binding affinities of small ligands by computer programs. Most drugs exert the action by binding to a receptor and this reaction is dictated by the free energy. Other interesting properties in drug design that could be investigated with free-energy methods are selectivity, transport properties and degradation. All of the methods presented in this thesis are based on the simulation of a model system and are either theoretically rigorous or of a more approximate nature. Some of the topic covered are How the precision and efficiency are affected by simulation protocols How the choice of energy model and solvation method affect the estimates How many unphysical states that needs to be simulated How small part of the receptor that is required in the model Estimates on a blind test of host–guest and protein–ligand affinities A benchmark of a rigorous method on 10 proteins and more than 100 ligands Together, these studies clearly show what methods to use and what methods to be more cautious about. It is concluded that approximate methods are generally not able predict very accurate affinities. Using rigorous methods, the absolute affinity of a molecule can be determined rather accurately but at considerable computational cost and with less precision. However, affinity differences between to very similar ligands can be accurately estimated both quickly and with a high precision. 1. It is necessary to average MM/GBSA energies over several independent simulations 2. It is possible to increase sampling by utalising different water boxes and alternative side-chain conformations 3. Semi-empirical QM methods can give accurate affinity estimates but require an empirical dispersion correction 4. Different polar solvation methods gives absolute affinities that can differ by 200 kJ/mol 5. No continuum non-polar solvation method can give accurate estimates on a wide range of systems because they neglect the microscopic detail of active-site water molecules 6. A truncated protein with a buffer region is a cost effective and accurate approach to obtain entropies 7. LIE is more efficient than MM/GBSA 8. It is hard to find an end-point continuum-solvent methods that generally give accurate results 9. MM/GBSA and LIE can be used to accurately predict host–guest affinities 10. Alchemical methods can be more efficient by a carefully design simulation protocol 11. The efficiency of alchemical methods can be improved by truncating the protein 12. Accurate relative estimates can be obtain from a single reference-state for a pairs of ligands with apolar substituents 13. Alchemical free-energy estimates converge much faster than MM/GBSA and entropy estimates 14. MM/PBSA is good at discriminating between binders and non-binders and fairly good at ranking binders 15. Accurate alchemical free-energies can be obtained for a large number of test cases A method to accurately estimate the binding affinity of a small molecule to a receptor would be indispensable in numerous fields. For instance, most drugs exert their action by binding to a macromolecule target. Thus, a lot of time and resources could be saved in drug design by predicting affinities by computer programs. In a series of 15 papers, we have tested, compared, and improved the most popular methods to estimate binding affinities. We have used for instance molecular mechanics with generalized Born and surface area solvation (MM/GBSA), linear interaction energy (LIE), and alchemical perturbation methods. Some of the topics covered are: ' How the precision of MM/GBSA estimates are affected by the simulation protocol ' If semiempirical quantum-mechanical methods can improve affinity estimates ' A comparison of different polar solvation methods in MM/GBSA ' If non-polar solvation methods can model different degrees of active-site hydration ' How to obtain normal-mode entropies accurately and efficiently ' What method is more efficient: LIE or MM/GBSA ' A comparison of several end-point continuum-solvation methods ' What charge model to use in simulations of host–guest complexes ' The performance of end-point methods in a binding-affinity blind test ' How we can make alchemical methods more useful for drug design ' If a single-reference state can be used to simulate several ligands ' What properties calculated from molecular dynamics simulations do converge Together, these studies clearly show what methods to use and what methods to avoid. We conclude that approximate methods are not very accurate and the results are highly system dependent. On the other hand, using alchemical methods, affinity differences between similar ligands can be accurately estimated both quickly and with a high precision.