Conformer Generation

If no experimentally determined structure of a compound is available or alternative conformations are needed, conformers of the molecule can be generated by computational means. A large number of conformer generation approaches have been developed for small organic molecules in the past decades. Two main search strategies are used to generate a conformational ensemble: systematic and stochastic. In the first approach, each rotatable bond is sampled systematically in discrete intervals, limiting the use to molecules with few rotatable bonds. Stochastic methods on the other hand sample the conformational space of a molecule randomly and can thus be applied to more flexible molecules. Recently, we have combined the stochastic-search method distance geometry, which is computationally efficient, with experimental torsion-angle preferences obtained from small molecule crystallographic data to improve the quality of the generated conformers [1]. The torsion angles were described by a series of hierarchically structured SMARTS patterns developed by Schärfer et al. [J. Med. Chem. 56, 2016 (2013)]. The new approach termed ETKDG has been implemented in the open-source cheminformatics library external pageRDKit and is freely available to the community.

[1] Riniker, Landrum, J. Chem. Inf. Model. (2015), 55, 2562.

As the first version of ETKDG relied on the SMARTS patterns for acyclic bonds developed by Schärfer et al., torsional sampling of aliphatic bonds in rings was not improved. For example, 6-membered aliphatic rings have a high preferences for chair conformations (i.e. all torsional angles around 60°). Recently, we developed new SMARTS patterns targeted for aliphatic ring bonds [2]. These resulted in torsional-angle distributions in the generated conformers, which better match the experimental distributions from crystal structures. This leads to a higher likelihood to generate a good ring conformation close to the crystal structure.

Due to the larger number of degrees of freedom of macrocycles, the conformational space to sample is much broader than for small molecules, creating a challenge for conformer generators. We therefore introduce into ETKDG different heuristics such as the usage of elliptical geometry and customizable Coulombic interactions to restrict the search space of macrocycles and bias the sampling toward more experimentally relevant structures [2]. In addition, bonds in macrocycles (defined here as rings with more than 8 bonds) were found to behave similar to acyclic bonds in terms of torsional-angle preferences, thus the same SMARTS patterns can be applied.

[2] external pageWang et al., J. Chem. Inf. Model. (2020), 60, 2044.

If experimental NMR data of a cyclic peptide is available, one can go a step further and incorporate this information directly in the conformer generation. For this, we extended the ETKDG conformer generator to include NOE-derived interproton distances in collaboration with scientists at Genentech [3]. NMR data from NOESY and ROESY experiments can easily be combined with distance geometry based conformer generators by modifying the molecular distance bounds matrix. In noeETKDG, the experimentally derived interproton distances are incorporated into the distance bounds matrix as loose upper (or lower) bounds to generate large conformer sets. Various sub-selection techniques can subsequently be applied to yield a conformer bundle that best reproduces the NOE data. The advantages of noeETKDG compared to other approaches are its speed and that no prior force-field parameterisation is required, which is especially useful for peptides with unnatural amino acids or synthetic macrocycles.

[3] external pageWang et al., J. Chem. Inf. Model. (2022), 62, 472.