End-user: BAM, DWA, Charite, Bayer

ZIBaffinity was designed to provide high-quality binding affinities ΔG of small chemical compounds to biological target molecules using atomistic molecular dynamics simulations on the basis of the Amber force field and methods of statistical thermodynamics.

1. Objectives and goals

Due to the mathematical complexity associated with the conformational space, the estimation of accurate biological host–guest (PL) binding affinities remains a difficult task that requires time-consuming atomistic force field simulations following complicated modeling and simulation protocols and requiring scripting skills. Nevertheless, they are useful in various scenarios comprising computational drug design and the toxicological risk assessment on anthropogenic substances as well as their biotic/abiotic transformation products (TP). These have been increasingly accumulating in various environmental compartments particularly including drinking water. An automated and easy-to-use workflow consisting of a reliable binding energy model, a database of critical pre-parameterized target proteins, and a cloud infrastructure with access to high performance computer facilities would substantially facilitate the calculation of binding energies for computational molecular modelling laymen.

2. How ZIBaffinity works

ΔG is calculated as the free energy difference between two distinct states of the ligand molecule, bound to the target and, respectively, free in water. Having uploaded a small drug-like molecule under observation as input, the user needs to select a potential protein target structure from a database of force field-parameterized 3D models. Using modern cloud orchestration technologies, the job is submitted to and processed on a high performance computer. Finally, the binding affinity along with few other thermodynamic quantities and a 3D model of the favorable protein-ligand binding mode (conformation) are then supplied to the user via email.

za reaction

Figure 1: Biological association dissociation reaction associated with Gibbs binding free energy ΔG.

Ensuing from the uploaded small molecule, numerous MD simulations including different starting positions of the molecular complex (distributed according to the icosahedron’s symmetry) and one unbound simulation are carried out simultaneously. The optimal binding mode (complex conformation) is extracted from that data and provided as a 3D molecular structure serving, along with thermo-statistical data as the basis for absolute or relative binding affinity estimation.

za complex

Figure 2: Natural hormone ligand 17-beta-estradiol bound to the human estrogen receptor alpha.

The affinity is estimated as a linear combination of averages of molecular observables according to an extended linear interaction energy model

za lie

where the parameter coefficients x_i need to be determined empirically in advance. Thus, the target data base provides only proteins, for which a training set of ligands with known binding affinities were available for the purpose of training these weights. In its urrent state the database consists only of the alpha estrogen receptor alpha as a proof-of-concept for which a highly reliable linear model has been developed.

za plot

Figure 3: Predicted/fitted versus experimentally determined binding energies for the estrogen receptor alpha.

3. Simulation Technology

  • Molecules parameterized according to Amber force field (Amber99sb & GAFF)

  • Estimation of ligand charges with the AM1BCC method

  • Explicit solvation through TIP4Pew water model

  • Deterministic molecular dynamics simulations of ligand molecule (bound to target and free in solution) using Gromacs

4. Features

  • Calculation of absolute binding free energies ΔG for biological host–guest systems

  • Calculation of unweighted thermodynamic contributions to ΔG in case of target molecules lacking training set

  • Determination of a favourable host–guest binding mode

5. Input/Output

  • Atomic Cartesian coordinates of a small chemical compound in MOL2 or PDB file format uploaded by user

  • One or more target molecules selected by user from a data base of parameterized proteins

  • Atomic coordinates of favourable protein-ligand binding mode output in PDB format

  • Either free energy of binding or, if no weights for linear model exist, the single energy terms

6. References

  • V. Durmaz, S. Schmidt, P. Sabri, C. Piechotta, M. Weber: A hands-off linear interaction energy approach to binding mode and affinity estimation of estrogens. 'Journal of Chemical Information and Modeling', 53:2681-2688, 2013.

  • V. Spahn, G. Del Vecchio, D. Labuz, A. Rodriguez-Gaztelumendi, N. Massaly, J. Temp, V. Durmaz, P. Sabri, M. Reidelbach, H. Machelska, M. Weber, C. Stein: A non-toxic pain killer designed by modeling of pathological receptor conformations. 'Science', 355:966-969, 2017.

  • K. Heye, D. Becker, C. Lütke-Eversloh, V. Durmaz, T. A. Ternes, M. Oetken, J. Oehlmann: Effects of carbamazepine and two of its metabolites on the non-biting midge Chironomus riparius in a sediment full life cycle toxicity test. 'Water Research', 98:19-72, 2016.