Ulf Ryde

Computational Chemistry, Lund University, Sweden - Research

Estimating ligand-binding affinities with quantum-mechanical methods

George Poulos, Meiting Wang, Martin A. Olsson, Octav Caldararu, Vilhelm Ekberg, Casper Steinmann, M. Misini Ignjatović, Pär Söderhjelm, Ulf Ryde*

Abstract

One of the largest challenges of computational chemistry is to estimate the binding free energy of a small molecule to a biomacromolecule (e.g. a drug candidate to its receptor). Currently, the best results are typically obtained by free-energy perturbation (FEP) methods, with free energies estimated by exponential averaging, thermodynamic integration or Bennett acceptance ratio [1]. Such methods require extensive sampling and therefore they have been mainly used with molecular-mechanics (MM) methods. However, it is well-known that such methods have severe limitations. Therefore, there have been quite some interest to improve binding-affinity estimates using quantum-mechanics (QM) methods [2]. We have tried to employ QM methods to in FEP estimates of binding affinities for both proteins and host–guest models. Initial attempts failed, because the perturbations did not converge [3, 4]. We have compared various reference-potential methods with explicit QM/MM FEP calculations [5]. They former are based on FEP calculations at the MM level, combined with MMQM/MM FEP calculations. We have tried to avoid QM/MM simulations by employing single-step exponential averaging or non-Boltzmann Bennett acceptance ratio method. For convergence, about 700 000 QM energy calculations were needed [6]. However, more reliable and accurate results are obtained with explicit QM/MM simulations [5], although they are very expensive. We have shown how they can be sped up by employing many short FEP calculations [7] or by use non-equilibrium simulations and Jarzynski’s equality [8]. These calculations still require millions of QM calculations. An alternative is to instead use structures minimized by QM/MM. We have developed and tested such methods for host–guest systems and proteins [9-11]. They are appreciably cheaper, but the accuracy is also worse.

References

1 A. S. J. S. Mey, B. K. Allen, H. E. B. Macdonald, J. D. Chodera, D. F. Hahn, M. Kuhn, J. Michel, D. L. Mobley, L. N. Naden, S. Prasad, A. Rizzi, J. Scheen, M. R. Shirts, G. Tresadern and H. Xu (2020) Living J Comp Mol Sci 2:18378-18378 2 U. Ryde and P. Söderhjelm (2016) Chem Rev 116:5520-5566 3 P. Mikulskis, D. Cioloboc, M. Andrejić, S. Khare, J. Brorsson, S. Genheden, R. A. Mata, P. Söderhjelm and U. Ryde (2014) J Comput-Aided Mol Design 28:375-400 4 S. Genheden, U. Ryde and P. Soderhjelm (2015) J Comput Chem 36:2114-2124 5 M. A. Olsson and U. Ryde (2017) J Chem Theory Comput 13:2245-2253 6 M. A. Olsson, P. Söderhjelm and U. Ryde (2016) J Comput Chem 37:1589-1600 7 C. Steinmann, M. A. Olsson and U. Ryde (2018) J Chem Theory Comput 14:3228-3237 8 M. Wang, Y. Mei and U. Ryde (2018) J Chem Theory Comput 14:6613-6622 9 O. Caldararu, M. A. Olsson, C. Riplinger, F. Neese and U. Ryde (2017) J Comput-Aided Mol Design 31:87-106 10 M. Misini Ignjatović, O. Caldararu, G. Dong, C. Muñoz-Gutierrez, F. Adasme-Carreño and U. Ryde (2016) J Comput-Aided Mol Design 30:707-730 11 O. Caldararu, M. A. Olsson, M. Misini Ignjatović, M. Wang and U. Ryde (2018) J Comput-Aided Mol Design 32:1027-1046