QM/MM free-energy perturbation and other methods to estimate ligand-binding affinities

Abstract: Experimental drug discovery is very time-consuming, risky and comes at a huge cost, typically several billion USD per drug. Even though decades of experimental drug discovery have provided cures of many diseases, there are still diseases for which there is no effective drug available. If drug development could be performedby theoretical and computational methods it would be of great advantage to humanity and is likely to accelerate drug discovery. One of the most promising computational methods is free-energy perturbation, which can provide estimates of protein–ligand binding affinities based on a molecular mechanics (MM)potential. Due to limitations of empirical potential-energy functions used to describe molecular interaction, there has been some interest to perform free-energy perturbation instead at the quantum-mechanics (QM)level of theory. To avoid the cost of performing sampling at the QM/MM level of theory, thermodynamic cycles can be employed. For this purpose, MM→QM free-energy perturbations in method space are required, but early applications have had convergence problems. In this thesis, different approaches toconverge QM/MM free-energy perturbations in method space are developed and compared to other methods to estimate protein–ligand binding affinities. Methods to obtain QM/MM energies by performing MM→QM free-energy perturbations using thermodynamic cycles are compared to direct alchemical free-energyperturbation with a QM/MM Hamiltonian. Moreover, alternative methods to improve free-energy perturbations at the MM level of theory by charge perturbations are assessed, as well as the use of QM/MM optimised structures. Furthermore, we study also the binding entropy contribution to ligand-binding affinities for thecancer target galectin-3. QM/MM free-energy perturbation calculations in this thesis have been converged to a precision of 1 kJ/mol. The calculated free energies agree with experimental data to within 4–6 kJ/mol, which allows for a proper ranking of lead candidates. For diastereomeric inhibitors of galectin-3, both qualitative and quantitative agreement between experimental and converged binding entropy contributions to binding affinities have been obtained with a precision of ~5 kJ/mol.