# CYP 2D6 Binding Affinity Predictions Using Multiple Ligand and Protein Conformations

^{1}

^{2}

^{*}

^{†}

## Abstract

**:**

## 1. Introduction

^{3}, which is larger than for CYPs 2A6, 2A13, 1A2 and 2E1, but smaller than for 2R1, 3A4, 2C8, and 2C9. Hritz et al. studied the impact of the plasticity and flexibility of CYP 2D6 on the accuracy of docking-based Site-of-Metabolism (SOM) prediction [14] and revealed that not only large conformational changes but also thermal motion can influence the reliability of structure-based SOM prediction. Interestingly, it was possible to select from MD simulations only a few CYP 2D6 structures that can be used as docking templates to obtain SOM prediction accuracies of more than 80% [14]. Later, positional fluctuations observed during MD [14] of side chains of active-site residues (Glu216, Phe483) could be confirmed by comparison to newly published X-ray structures (PDB codes 3QM4 [15], 3TDA and 3TBG) [22].

## 2. Iterative LIE approach

_{i}for the ligand binding in pose i to the protein is then calculated as

_{i}of an independent simulation i to the interaction energy of the protein-bound ligand with its surrounding can be calculated [17] as

_{calc}of that ligand averaged over the i independent simulations can be calculated from

_{i}’s are obtained by applying an iterative scheme, as described by Stjernschantz and Oostenbrink [17]. After an initial guess, the LIE coefficients α and β are iteratively optimized until convergence is reached, i.e., by obtaining a minimum value for the root-mean-square error between calculated and experimental values for the binding free energies. Relative experimental binding free energies ΔG

_{exp}for the considered thioureas were derived from inhibition data reported by Onderwater [24] (Table 1).

## 3. Results and Discussion

_{calc}) and experimental binding free energies (ΔG

_{exp}). Tables 2 and 3 report α, β and RMSE values for the different LIE models as obtained for the S1 and S2 sets of simulations, together with errors in the prediction for the compounds (outliers) for which ΔG

_{calc}deviates by more than 1 kcal mol

^{−}

^{1 (}4.184 kJ mol

^{−}

^{1}) from ΔG

_{exp}.

_{calc}for only one (S1) or five (S2) of the ligands (see Supplementary Material, Table S1).

_{bind}from multiple MD simulations per ligand requires that individual simulations cover different parts of protein-ligand conformational space. For several ligands, different hydrogen bond interactions with the protein were observed when comparing pairs of S1 and S2 simulations starting from the same protein-ligand conformation (see Supplementary Material, Table S2 and Figure S1). This can be seen as a motivation to treat S1 and S2 simulations individually, as in the calibration of the S1–S2 models. On the other hand, relatively short MD simulations were performed in this study, justified by the fact that conformational sampling is already partly achieved by including results from simulations starting from different protein conformations and ligand-binding orientations. Therefore, large conformational changes cannot be expected during a single simulation and based on this argument, pairs of S1 and S2 simulations starting from a single configuration should not be assigned separate weights W

_{i}in calibrating the LIE models. In the S1-S2-A models presented in Table 5, 〈V

_{EL}〉 and 〈V

_{V dW}〉 obtained from pairs of S1 and S2 simulations (starting from the same protein-ligand conformation) were averaged for use in Equation (3). Note that in the limit of infinite sampling, the ensemble averages of the interaction energies would be identical for simulations S1 and S2, leading to identical weights (Equation 2) and the difference between models S1-S2 and S1-S2-A disappears. From the RMSEs of the S1-S2 and S1-S2-A models (Tables 4 and 5) and the obtained correlation between calculated and experimental binding affinities (Figure 2), we found similar performance in binding free energy prediction when using either one of the schemes, despite of the slight differences in calibrated β values (Tables 4 and 5). In addition, the S1–S2 and S1-S2-A sets demonstrate a similar profile in terms of the dependence of the model’s RMSE on the calibrated α and β values (Figure 3), which also shows a larger sensitivity of the RMSE on the α than on the β parameter. This indicates a larger dependency of the predicted binding free energy on differences in (apolar) van der Waals interactions than on changes in electrostatic interactions between ligand and environment upon binding, in line with an earlier LIE model for CYP 1A2 as presented by Vasanthanathan et al. [25]. For CYP 1A2, Vasanthanathan found that the predictive accuracy of the free-energy models improved upon forcing β to zero in Equation (1) and introducing a constant γ as additional off-set parameter. Here, we found for some of the models in Tables 2–5 a small decrease in RMSE when recalibrating our models using α and γ as fitting parameters (instead of α and β). For example, for the models in Tables 2–5 in which β was found to adopt an unphysical (negative) value, RMSEs were found to decrease by 0.15 (M1/S1), 0.66 (P170/S2), 0.03 (P170/S1-S2) and 0.14 kJ mol

^{−}

^{1}(P170/S1-S2-A), respectively. However, when fitting a model with α and γ as parameters and using results from all simulations, we found an increase in RMSE by 0.62 kJ mol

^{−}

^{1}(S1–S2) and 0.72 kJ mol

^{−}

^{1}(S1-S2-A) when compared to values in Tables 4 and 5, indicating that the dependency of the binding free energy on electrostatic interactions should not be neglected in these cases.

^{−}

^{1}, Table 5), but when including simulations starting from the M1 poses as well, a β value was obtained that is in turn slightly closer to the theoretical value of 0.5 [9]. Moreover, when inspecting contributions from the sets of simulations included in the S1-S2 and S1-S2-A LIE models based on all MD runs (Tables 6 and 7), significant contributions to the predicted free energies were not only obtained from simulations starting from the M2 pose, but also from those starting from M1 (with the latter even dominating for most of the ligands in case of S1-S2-A, Table 7).

^{−}

^{1}, Table 5). On the other hand, when comparing the LIE models based on all simulations with the P170 models, improvement is also observed in terms of a significant increase in β. In addition, both protein conformations contribute significantly to the predicted binding free energies (Tables 6 and 7). These findings demonstrate the importance for the flexible Cytochrome P450 2D6 enzyme of including MD simulations starting from different protein-ligand conformations to obtain accurate binding affinities. Moreover, because sampling of protein-ligand conformational space is already partly accounted for by combining MD simulations starting from distinct configurations, relatively short MD simulations (1 ns) are sufficient to obtain RMSE errors well below 1 kcal mol

^{−}

^{1}. In contrast, calculation of relative binding free energy differences between any pair of ligands by using rigorous free energy methods such as thermodynamic integration [8] or free-energy perturbation [7] would probably require a series of simulations on the nanosecond time scale for every perturbation of a given ligand into another one. In summary, the chosen approach does not only improve the predictive quality of the method, but also its computational attractability and efficiency. Because our approach relies on averaging over multiple independent simulations, its efficiency can be optimized by using implementations of MD software on massively parallel computing facilities that are available within e.g., the Folding@Home [26] andWeNMR communities [27]. An additional strength of the chosen approach is that a priori, no knowledge is required of the dominant protein configuration. In a follow-up study [23] we will show how our approach can be automated (e.g., for industrial application) by developing a LIE model for a different class of CYP 2D6 binders, for which pre-selection of possible ligand-binding modes (based on visual inspection of docked ligand poses) is not feasible.

## 4. Computational Methods

#### 4.1. Choice of the Protein Coordinate Templates

_{1}of Phe483 adopting a value of approximately 70°, whereas the docking template used for large compounds exhibits a χ

_{1}value of 170°. To account for this structural variety, we started our docking and LIE/MD studies from two CYP 2D6 structures that were identified by Hritz to give maximum accuracy in docking-based SOM predictions, and which differ in the Phe483 χ

_{1}values. In the current work, P70 (χ

_{1}≅ 70°) refers to the first of these structures (denoted as PPD-70-fr-216 in the Hritz paper), and the second one (referred to as PPD-170-fr-173 in the Hritz paper) is denoted here as P170 (χ

_{1}≅ 170°).

#### 4.2. Docking

_{ε}atom toward the heme iron (which was chosen to be maximally 0.6 nm [32]) and (ii) the orientation of the imidazole ring with respect to the heme group (Figure 4). In the first of the selected poses (denoted as M1), the imidazole-C

_{δ}atom of the ligand is directed towards the heme carboxyl groups (see e.g., Figure 4a,c). In the second selected pose (denoted as M2), C

_{δ}is directed in the opposite direction (Figure 4b,d). By combining the two different protein structures (P70 in Figure 4a,b and P170 in Figure 4c,d) and ligand binding modes (M1 and M2), four different protein-ligand complex conformations were chosen per ligand to start MD simulations from.

#### 4.3. Molecular Dynamics Simulation

^{+}counter ions (ligand in protein), in order to neutralize the net charge of the protein-ligand complex, after which the system was energy minimized again. During MD simulations, the equations of motion were integrated using the leap-frog scheme [36] with a timestep of 2 fs. The SHAKE algorithm [37] was applied to constrain all bond lengths to their zero-energy value, using a relative geometric tolerance of 10

^{−}

^{4}. The value of the χ

_{1}dihedral angle of protein residue Phe483 was restrained to either 70° or 170° using a harmonic potential with a force constant of 30.0 kJ mol

^{−}

^{1}deg

^{−}

^{2}, in order to prevent possible transitions between P70 and P170 protein conformations during simulation and to ensure that different parts of protein-ligand conformational space were sampled in independent MD simulations (which is a prerequisite for use of Equation (2) to calculate their relative weights [17]). During MD, the temperature was maintained close to its reference value (300 K) by weakly coupling the solute and solvent degrees of freedom separately to a heat bath [38], with a relaxation time of 0.1 ps. The pressure was maintained close to its reference value (1 atm) by weak coupling of the particle coordinates and box dimensions (isotropic coordinate scaling) to a pressure bath [38], using a relaxation time of 0.5 ps and an isothermal compressibility of 0.4575 .10

^{−}

^{3}kJ

^{−}

^{1}mol nm

^{3}. Non-bonded interactions were computed using a twin-range scheme [39,40], with short- and long-range cutoff distances set to 0.8 and 1.4 nm, respectively, and a frequency of 5 time steps for the update of the short-range pairlist and for the calculation of intermediate-range interactions. A reaction-field correction [41] was applied to account for the mean effect of omitted electrostatic interactions beyond the long-range cutoff distance, using a relative dielectric permittivity of 61, which is appropriate for the SPC water model [42]. During the first 100 ps of equilibration, the temperature was increased in a stepwise manner up to 300 K, by increasing the temperature by 60 K every 20 ps and by applying harmonic positional restraints to the protein backbone atoms, using a force constant decreasing from 25,000 kJ mol

^{−}

^{1}nm

^{−}

^{1}(at 60 K) to 25 kJ mol

^{−}

^{1}nm

^{−}

^{1 (}at 240 K). After an equilibration period of in total 2.1 ns, all simulations were carried out for a duration of 1 ns during which interaction energies and coordinates were stored every 0.02 and 0.4 ps, respectively, for subsequent analysis. Starting from all (four) combinations of protein conformations (P70 and P170) and ligand-binding orientations (M1 and M2), two independent MD simulations were performed (designated as S1 and S2, respectively). For each pair of S1 and S2 simulations, the same starting conformation of the protein-thiourea complex was used, but different (random) sets of atomic velocities were assigned at the beginning of the thermal equilibration phase.

## 5. Conclusions

^{−}

^{1}. The applied LIE approach serves as a promising template to develop efficient ligand-binding affinity prediction tools for very flexible and malleable proteins such as CYPs.

## Supplementary file

ijms-14-24514-s001.pdf## Acknowledgements

## Conflicts of Interest

## References

- Christ, C.D.; Mark, A.E.; van Gunsteren, W.F. Basic ingedients of free energy calculations: A review. J. Comp. Chem
**2010**, 31, 1569–1582. [Google Scholar] - De Ruiter, A.; Oostenbrink, C. Free energy calculations of protein-ligand interactions. Curr. Opin. Chem. Biol
**2011**, 15, 547–552. [Google Scholar] - Ferrara, P.; Gohlke, H.; Price, D.J.; Klebe, G.; Brooks, C.L., III. Assessing scoring functions for protein-ligand interactions. J. Med. Chem
**2004**, 47, 3032–3047. [Google Scholar] - Warren, G.L.; Andrews, C.W.; Capelli, A.M.; Clarke, B.; LaLonde, J.; Lambert, M.H.; Lindvall, M.; Nevins, N.; Semus, S.F.; Senger, S.; et al. A critical assessment of docking programs and scoring functions. J. Med. Chem
**2005**, 49, 5912–5931. [Google Scholar] - Stjernschantz, E.; Marelius, J.; Medina, C.; Jacobsson, M.; Vermeulen, N.P.E.; Oostenbrink, C. Are automated molecular dynamics simulations and binding free energy calculations realistic tools in lead optimization? An evaluation of the linear interaction energy (LIE) method. J. Chem. Inf. Model
**2006**, 46, 1972–1983. [Google Scholar] - Cheng, T.; Li, X.; Li, Y.; Liu, Z.; Wang, R. Comparative assessment of scoring functions on a diverse test set. J. Chem. Inf. Model
**2009**, 49, 1079–1093. [Google Scholar] - Zwanzig, R.W. High-temperature equation of state by a perturbation method. I. Nonpolar gases. J. Chem. Phys
**1954**, 22, 1420–1426. [Google Scholar] - Beveridge, D.L.; DiCapua, F.M. Free-energy via molecular simulation—Applications to chemical and biomolecular systems. Annu. Rev. Biophys. Biophys. Chem
**1989**, 18, 431–492. [Google Scholar] - Åqvist, J.; Medina, C. A new method for predicting binding affinity in computer-aided drug design. Protein Eng
**1994**, 7, 385–391. [Google Scholar] - Hansson, T.; Marelius, J.; Åqvist, J. Ligand binding affinity prediction by linear interaction energy methods. J. Comput.-Aided Mol. Des
**1998**, 12, 27–35. [Google Scholar] - Guengerich, F.P. A malleable catalyst dominates the metabolism of drugs. Proc. Natl. Acad. Sci. USA
**2006**, 103, 13565–13566. [Google Scholar] - Oostenbrink, C.; de Ruiter, A.; Hritz, J.; Vermeulen, N.P.E. Malleability and versatility of cytochrome P450 active sites studied by molecular simulations. Curr. Drug Metab
**2012**, 13, 190–196. [Google Scholar] - Ekroos, M.; Sjögren, T. Structural basis for ligand promiscuity in cytochrome P450 3A4. Proc. Natl. Acad. Sci. USA
**2006**, 103, 13682–13687. [Google Scholar] - Hritz, J.; de Ruiter, A.; Oostenbrink, C. Impact of plasticity and flexibility on docking results for cytochrome P450 2D6: A combined approach of molecular dynamics and ligand docking. J. Med. Chem
**2008**, 51, 7469–7477. [Google Scholar] - Wang, A.; Savas, U.; Hsu, M.H.; Stout, C.D.; Johnson, E.F. Crystal structure of Human Cytochrome P450 2D6 with Prinomastat Bound. J. Biol. Chem
**2012**, 287, 10834–10843. [Google Scholar] - Pochapsky, T.C.; Kazanis, S.; Dang, M. Conformational plasticity and structure/function relationships in Cytochromes P450. Antioxid. Redox Signal
**2010**, 13, 1273–1296. [Google Scholar] - Stjernschantz, E.; Oostenbrink, C. Improved ligand-protein binding affinity predictions using multiple binding modes. Biophys. J
**2010**, 98, 2682–2691. [Google Scholar] - Faver, J.C.; Yang, W.; Merz, K.M., Jr. The Effects of computational modeling errors on the estimation of statistical mechanical variables. J. Chem. Theo. Comput
**2012**, 8, 3769–3776. [Google Scholar] - Williams, J.A.; Hyland, R.; Jones, B.C.; Smith, D.A.; Hurst, S.; Goosen, T.C.; Peterkin, V.; Koup, J.R.; Ball, S.E. Drug-drug interactions for UDP-glucuronosyltransferase substrates: A pharmacokinetic explanation for typically observed low exposure (AUC(i)/AUC) ratios. Drug Metab. Dispos
**2004**, 32, 1201–1208. [Google Scholar] - Bazeley, P.S.; Prithivi, S.; Struble, C.A.; Povinelli, R.J.; Sem, D.S. Synergistic use of compound properties and docking scores in neural network modeling of CYP2D6 binding: Predicting affinity and conformational sampling. J. Chem. Inf. Model
**2006**, 46, 2698–2708. [Google Scholar] - Rowland, P.; Blaney, F.E.; Smyth, M.G.; Jones, J.J.; Leydon, V.R.; Oxbrow, A.K.; Lewis, C.J.; Tennant, M.G.; Modi, S.; Eggleston, D.S.; Chenery, R.J.; Bridges, A.M. Crystal structure of human cytochrome P450 2D6. J. Biol. Chem
**2006**, 281, 7614–7622. [Google Scholar] - Oostenbrink, C. Structure-based Methods for Predicting the Sites and Products of Metabolism. In Drug Metabolism Prediction; Kirchmair, J., Ed.; Wiley: Weinheim, Germany, 2013. [Google Scholar]
- Vosmeer, C.R.; Pool, R.; van Stee, M.F.; Vermeulen, N.P.E.; Geerke, D.P. Towards automated binding affinity prediction using an iterative linear interaction energy approach. Int. J. Mol. Sci
**2013**. Submitted for publication. [Google Scholar] - Onderwater, R. Molecular Toxicology of Thiourea-containing Compounds. Ph.D. Thesis, Vrije Universiteit, Amsterdam, The Netherlands, 2005. [Google Scholar]
- Vasanthanathan, P.; Olsen, L.; Jørgensen, F.S.; Vermeulen, N.P.E.; Oostenbrink, C. Computational prediction of binding affinity for CYP1A2-ligand complexes using empirical free energy calculations. Drug Metab. Dispos
**2010**, 38, 1347–1354. [Google Scholar] - Pande, V.S.; Baker, I.; Chapman, J.; Elmer, S.P.; Khaliq, S.; Larson, S.M.; Rhee, Y.M.; Shirts, M.R.; Snow, C.D.; Sorin, E.J.; Zagrovic, B. Atomistic protein folding simulations on the submillisecond time scale using worldwide distributed computing. Biopolymers
**2003**, 68, 91–109. [Google Scholar] - Van Dijk, M.; Wassenaar, T.A.; Bonvin, A.M.J.J. A flexible, grid-enabled web portal for GROMACS molecular dynamics simulations. J. Chem. Theory Comput
**2012**, 8, 3463–3472. [Google Scholar] - Jones, G.; Willett, P.; Glen, R.C.; Leach, A.R.; Taylor, R.D. Development and validation of a genetic algorithm for flexible docking. J. Mol. Biol
**1997**, 267, 727–748. [Google Scholar] - Eldridge, M.D.; Murray, C.W.; Auton, T.R.; Paolini, G.V.; Mee, R.P. Binding affinity prediction; Empirical scoring function; De novo molecular design; Protein-ligand complexes. J. Comput. Aided Mol. Des
**1997**, 11, 425–445. [Google Scholar] - Kirton, S.B.; Murray, C.W.; Verdonk, M.L.; Taylor, R.D. Prediction of binding modes for ligands in the Cytochromes P450 and the other heme-containing proteins. Proteins
**2005**, 58, 836–844. [Google Scholar] - Santos, R.; Hritz, J.; Oostenbrink, C. Role of water in molecular docking simulations of cytochrome P450 2D6. J. Chem. Inf. Model
**2010**, 50, 146–154. [Google Scholar] - De Graaf, C.; Oostenbrink, C.; Keizers, P.H.J.; van der Wijst, T.; Jongejan, A.; Vermeulen, N.P.E. Catalytic site prediction and virtual screening of Cytochrome P450 2D6 substrates by consideration of water and rescoring in automated docking. J. Med. Chem
**2006**, 20, 2417–2430. [Google Scholar] - Schmid, N.; Christ, C.D.; Christen, M.; Eichenberger, A.P.; van Gunsteren, W.F. Architecture, implementation and parallelisation of the GROMOS software for biomolecular simulation. Comp. Phys. Commun
**2012**, 183, 890–903. [Google Scholar] - Lins, R.D.; Hünenberger, P.H. A new GROMOS force field for hexopyranose-based carbohydrates. J. Comp. Chem
**2005**, 26, 1400–1412. [Google Scholar] - Berendsen, H.J.C.; Postma, J.P.M.; van Gunsteren, W.F.; Hermans, J. Intermolecular Forces; Pullman, B., Ed.; Reidel: Dordrecht, The Netherlands, 1981; pp. 331–338. [Google Scholar]
- Hockney, R.W. The potential calculation and some applications. Methods Comput. Phys
**1970**, 9, 136–211. [Google Scholar] - Ryckaert, J.-P.; Ciccotti, G.; Berendsen, H.J.C. Numerical integration of the cartesian equations of motion of a system with constraints: Molecular dynamics of n-alkanes. J. Comput. Phys
**1977**, 23, 327–341. [Google Scholar] - Berendsen, H.J.C.; Postma, J.P.M.; van Gunsteren, W.F.; Di Nola, A.; Haak, J.R. Molecular-dynamics with coupling to an external bath. J. Chem. Phys
**1984**, 81, 3684–3690. [Google Scholar] - Van Gunsteren, W.F.; Berendsen, H.J.C. Computer simulation of molecular dynamics: Methodology, applications and perspectives in chemistry. Angew. Chem. Int. Ed. Engl
**1990**, 29, 992–1023. [Google Scholar] - Van Gunsteren, W.F.; Billeter, S.R.; Eising, A.A.; Hünenberger, P.H.; Krüger, P.; Mark, A.E.; Scott, W.R.P.; Tironi, I.G. Biomolecular Simulation: The GROMOS96 Manual and User Guide; vdf Hochschulverlag: ETH Zurich, Switzerland, 1996. [Google Scholar]
- Tironi, I.G.; Sperb, R.; Smith, P.E.; van Gunsteren, W.F. A generalized reaction field method for molecular dynamics simulations. J. Chem. Phys
**1995**, 102, 54541–5495. [Google Scholar] - Heinz, T.N.; van Gunsteren, W.F.; Hünenberger, P.H. Comparison of four methods to compute the dielectric permittivity of liquids from molecular dynamics simulations. J. Chem. Phys
**2001**, 115, 1125–1136. [Google Scholar]

**Figure 2.**Correlation between calculated (ΔG

_{calc}) and experimental (ΔG

_{exp}) binding free energies obtained for LIE models S1–S2 (

**a**) and S1-S2-A (

**b**), which combine results from all MD simulations. The solid line indicates ideal correlation between ΔG

_{exp}and ΔG

_{calc}, and thin dashed lines represent errors of ±4.184 kJ mol

^{−}

^{1}(1 kcal mol

^{−}

^{1}).

**Figure 3.**3D contour plots of the dependency of the RMSE (root-mean square error between calculated and experimental binding free energies) on α and β values for the S1–S2 (

**a**) and S1-S2-A (

**b**) LIE models, which are based on inclusion of results from all MD simulations. The color bar indicates a range of RMSE values between 2 and 5 kJ mol

^{−}

^{1}. RMSE values of 5 kJ mol

^{−}

^{1}and higher are depicted in yellow.

**Figure 4.**Typical ligand-binding orientations to start MD simulations from: (

**a**) pose P70-M1; (

**b**) pose P70-M2; (

**c**) pose P170-M1; and (

**d**) pose P170-M2. Residue Phe483, ligand (L1) and the CYP heme group are shown.

**Table 1.**Experimental binding free energies ΔG

_{exp}for the thiourea compounds considered, derived from IC

_{50}values reported in a CYP 2D6 inhibition study with 7-methoxy-4(aminomethyl)coumarin (MAMC) as a fluorometric probe [24].

LIGAND | R | IC_{50}(μM) | ΔG_{exp}(kJ mol^{−1}) |
---|---|---|---|

L1 | Ethyl | 87 ± 21 | −23.32 ± 0.69 |

L2 | 1-Propyl | 42 ± 14 | −25.14 ± 1.01 |

L3 | cyclo-Hexyl | 34 ± 15 | −25.66 ± 1.45 |

L4 | Phenyl | 57 ± 17 | −24.37 ± 0.88 |

L5 | p-Methylphenyl | 17 ± 6.0 | −27.39 ± 1.09 |

L6 | p-Methoxyphenyl | 21 ± 8.7 | −26.86 ± 1.34 |

L7 | p-Chlorophenyl | 3.5 ± 1.2 | −31.33 ± 1.05 |

L8 | Methylphenyl | 7.0 ± 1.2 | −29.61 ± 0.47 |

L9 | Methyl-(p-methoxy)phenyl | 4.1 ± 1.4 | −30.94 ± 1.04 |

L10 | Ethylphenyl | 0.60 ± 0.13 | −35.73 ± 0.61 |

**Table 2.**α and β parameters for the LIE models obtained from simulations S1. For each model, root-mean-square errors (RMSEs, in kJ mol

^{−}

^{1}) between calculated (ΔG

_{calc}) and experimental binding free energies (ΔG

_{exp}) are shown as well, together with errors (in kJ mol

^{−}

^{1}) in the prediction for the compounds (outliers) for which ΔG

_{calc}deviates by more than 1 kcal mol

^{−}

^{1}(4.184 kJ mol

^{−}

^{1}) from ΔG

_{exp}.

Runs | MODEL | β | α | RMSE | L1 | L2 | L3 | L4 | L5 | L6 | L7 | L8 | L9 | L10 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

S1 | P70-M1 | 0.239 | 0.425 | 3.65 | 5.01 | 7.69 | 4.77 | |||||||

P70-M2 | 0.142 | 0.434 | 1.39 | |||||||||||

P170-M1 | 0.084 | 0.414 | 4.07 | 5.46 | 5.18 | 6.78 | ||||||||

P170-M2 | 0.231 | 0.493 | 4.03 | 6.43 | 6.39 | 6.45 | ||||||||

P70 | 0.223 | 0.416 | 2.95 | 7.03 | 4.26 | |||||||||

P170 | 0.066 | 0.398 | 3.68 | 4.24 | 4.98 | 4.61 | 5.30 | 4.33 | ||||||

M1 | −0.142 | 0.301 | 3.17 | 6.69 | 6.17 | |||||||||

M2 | 0.174 | 0.437 | 1.40 | |||||||||||

all | 0.219 | 0.415 | 2.78 | 6.4 | 4.82 |

**Table 3.**α and β parameters for the LIE models obtained from simulations S2. For each model, root-mean-square errors (RMSEs, in kJ mol

^{−}

^{1}) between calculated (ΔG

_{calc}) and experimental binding free energies (ΔG

_{exp}) are shown as well, together with errors (in kJ mol

^{−}

^{1}) in the prediction for the compounds (outliers) for which ΔG

_{calc}deviates by more than 1 kcal mol

^{−}

^{1}(4.184 kJ mol

^{−}

^{1}) from ΔG

_{exp}.

Runs | MODEL | β | α | RMSE | L1 | L2 | L3 | L4 | L5 | L6 | L7 | L8 | L9 | L10 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

S2 | P70-M1 | 0.153 | 0.400 | 2.70 | 4.19 | 5.50 | ||||||||

P70-M2 | 0.119 | 0.405 | 4.61 | 4.20 | 6.33 | 8.49 | 6.78 | 5.42 | ||||||

P170-M1 | 0.048 | 0.407 | 4.26 | 11.35 | ||||||||||

P170-M2 | 0.031 | 0.412 | 5.56 | 7.11 | 11.10 | 10.65 | ||||||||

P70 | 0.176 | 0.404 | 3.38 | 6.32 | 5.73 | |||||||||

P170 | −0.069 | 0.351 | 4.28 | 4.52 | 11.10 | |||||||||

M1 | 0.116 | 0.392 | 2.60 | 5.29 | ||||||||||

M2 | 0.107 | 0.395 | 3.80 | 5.65 | 7.17 | 6.39 | ||||||||

all | 0.146 | 0.397 | 3.18 | 5.70 | 5.98 |

**Table 4.**α and β parameters for the LIE models obtained from simulation sets S1 and S2, by considering S1 and S2 sets separately in Equation (3). For each LIE model, root-mean-square errors (RMSEs, in kJ mol

^{−}

^{1}) between calculated (ΔG

_{calc}) and experimental binding free energies (ΔG

_{exp}) are shown as well, together with errors (in kJ mol

^{−}

^{1}) in the prediction for the compounds (outliers) for which ΔG

_{calc}deviates by more than 1 kcal mol

^{−}

^{1}(4.184 kJ mol

^{−}

^{1}) from ΔG

_{exp}.

Runs | MODEL | β | α | RMSE | L1 | L2 | L3 | L4 | L5 | L6 | L7 | L8 | L9 | L10 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

S1–S2 | P70-M1 | 0.109 | 0.381 | 3.15 | 5.24 | 6.47 | ||||||||

P70-M2 | 0.120 | 0.400 | 3.70 | 5.45 | 7.66 | 4.53 | ||||||||

P170-M1 | 0.030 | 0.381 | 3.19 | 4.25 | 4.84 | 4.79 | 4.61 | |||||||

P170-M2 | 0.010 | 0.394 | 4.79 | 5.33 | 7.12 | 10.72 | ||||||||

P70 | 0.215 | 0.411 | 2.92 | 6.37 | 4.30 | |||||||||

P170 | −0.056 | 0.343 | 3.11 | 4.24 | 5.93 | |||||||||

M1 | 0.063 | 0.366 | 3.02 | 6.49 | ||||||||||

M2 | 0.162 | 0.406 | 3.22 | 6.63 | 4.26 | 6.39 | ||||||||

all | 0.217 | 0.411 | 2.55 | 5.75 |

**Table 5.**α and β parameters for the LIE models obtained from simulation sets S1 and S2, by averaging results for pairs of S1 and S2 simulations. For each LIE model, root-mean-square errors (RMSEs, in kJ mol

^{−}

^{1}) between calculated (ΔG

_{calc}) and experimental binding free energies (ΔG

_{exp}) are shown as well, together with errors (in kJ mol

^{−}

^{1}) in the prediction for the compounds (outliers) for which ΔG

_{calc}deviates by more than 1 kcal mol

^{−}

^{1}(4.184 kJ mol

^{−1}) from ΔG

_{exp}.

Runs | MODEL | β | α | RMSE | L1 | L2 | L3 | L4 | L5 | L6 | L7 | L8 | L9 | L10 |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

S1-S2-A | P70-M1 | 0.146 | 0.401 | 3.09 | 5.78 | 6.11 | ||||||||

P70-M2 | 0.149 | 0.428 | 2.78 | 5.62 | ||||||||||

P170-M1 | 0.017 | 0.400 | 3.34 | 5.39 | 4.53 | 6.75 | ||||||||

P170-M2 | 0.142 | 0.455 | 4.58 | 4.42 | 7.30 | 10.18 | ||||||||

P70 | 0.203 | 0.419 | 2.78 | 5.90 | ||||||||||

P170 | −0.000 | 0.389 | 3.45 | 4.54 | 7.87 | |||||||||

M1 | 0.108 | 0.395 | 3.13 | 4.97 | 6.89 | |||||||||

M2 | 0.170 | 0.430 | 2.33 | 4.29 | ||||||||||

all | 0.190 | 0.419 | 2.59 | 5.50 |

**Table 6.**Relative weights W

_{i}of the different simulations i to the binding free energies calculated for ligands L1–L10 using the S1–S2 LIE model (as obtained by combining results from all simulations).

Ligand | P70-M1-S1 | P70-M2-S1 | P170-M1-S1 | P170-M2-S1 | P70-M1-S2 | P70-M2-S2 | P170-M1-S2 | P170-M2-S2 |
---|---|---|---|---|---|---|---|---|

L1 | 0.026 | 0.165 | 0.218 | 0.063 | 0.101 | 0.011 | 0.030 | 0.386 |

L2 | 0.091 | 0.047 | 0.029 | 0.018 | 0.120 | 0.630 | 0.053 | 0.013 |

L3 | 0.366 | 0.004 | 0.059 | 0.257 | 0.213 | 0.028 | 0.025 | 0.047 |

L4 | 0.157 | 0.014 | 0.392 | 0.145 | 0.113 | 0.034 | 0.113 | 0.031 |

L5 | 0.371 | 0.025 | 0.057 | 0.075 | 0.166 | 0.014 | 0.286 | 0.006 |

L6 | 0.268 | 0.004 | 0.118 | 0.001 | 0.113 | 0.261 | 0.116 | 0.119 |

L7 | 0.225 | 0.168 | 0.022 | 0.003 | 0.185 | 0.391 | 0.002 | 0.005 |

L8 | 0.150 | 0.113 | 0.008 | 0.027 | 0.206 | 0.046 | 0.309 | 0.140 |

L9 | 0.078 | 0.019 | 0.313 | 0.024 | 0.199 | 0.060 | 0.064 | 0.243 |

L10 | 0.098 | 0.621 | 0.056 | 0.010 | 0.074 | 0.136 | 0.003 | 0.001 |

**Table 7.**Relative weights W

_{i}of the different simulations i to the binding free energies calculated for ligands L1-L10 using the S1-S2-A LIE model (as obtained by combining results from all simulations).

Ligand | P70-M1 | P70-M2 | P170-M1 | P170-M2 |
---|---|---|---|---|

L1 | 0.177 | 0.150 | 0.266 | 0.408 |

L2 | 0.309 | 0.538 | 0.108 | 0.044 |

L3 | 0.608 | 0.038 | 0.086 | 0.267 |

L4 | 0.293 | 0.068 | 0.483 | 0.156 |

L5 | 0.663 | 0.063 | 0.219 | 0.055 |

L6 | 0.556 | 0.137 | 0.264 | 0.043 |

L7 | 0.515 | 0.461 | 0.017 | 0.007 |

L8 | 0.473 | 0.189 | 0.146 | 0.192 |

L9 | 0.341 | 0.085 | 0.387 | 0.187 |

L10 | 0.257 | 0.687 | 0.047 | 0.009 |

© 2013 by the authors; licensee MDPI, Basel, Switzerland This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

## Share and Cite

**MDPI and ACS Style**

Perić-Hassler, L.; Stjernschantz, E.; Oostenbrink, C.; Geerke, D.P.
CYP 2D6 Binding Affinity Predictions Using Multiple Ligand and Protein Conformations. *Int. J. Mol. Sci.* **2013**, *14*, 24514-24530.
https://doi.org/10.3390/ijms141224514

**AMA Style**

Perić-Hassler L, Stjernschantz E, Oostenbrink C, Geerke DP.
CYP 2D6 Binding Affinity Predictions Using Multiple Ligand and Protein Conformations. *International Journal of Molecular Sciences*. 2013; 14(12):24514-24530.
https://doi.org/10.3390/ijms141224514

**Chicago/Turabian Style**

Perić-Hassler, Lovorka, Eva Stjernschantz, Chris Oostenbrink, and Daan P. Geerke.
2013. "CYP 2D6 Binding Affinity Predictions Using Multiple Ligand and Protein Conformations" *International Journal of Molecular Sciences* 14, no. 12: 24514-24530.
https://doi.org/10.3390/ijms141224514