Shuzhe Wang Abstract

Improving RDKit’s Conformer Generator to Sample Macrocycles

Shuzhe Wang¹, Sereina Riniker¹

¹ETH Zürich

Macrocycles are a novel class of drug candidates gaining increased attention. Compared to traditional small-molecule drugs, their bigger sizes enable them to bind to large but flat binding sites with high affinity. This opens doors for new therapeutic targets like protein-protein interfaces (PPIs) and GPCRs – targets that are often considered to be undruggable[1]. A well-sampled conformational ensemble of a macrocycle will facilitate many types of in silico studies, such as MD and 3D-QSAR or even QM-based calculations. However, 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. One approach to address this issue is to restrict the search space of macrocycles, in particular we are interested in biasing the generator towards conformers with higher internal order (usually manifested in intramolecular hydrogen bonding).

Here, we present an approach by adapting the existing code in the cheminformatics package RDKit – the ETKDG conformer generator [2] – a distance geometry based method designed for small molecule geometries. The ETKDG workflow starts with calculating the molecular distance bounds matrix, followed by generating atom coordinates satisfying the lower and upper bounds, and subsequent structural refinement via various force terms (e.g. empirical torsion potentials). Our adaptations for macrocycles are introduced at three points in the workflow:

1. The behaviour of amides in large rings is modified to have a high preference for the trans configuration.

2. The concept of eccentricity from 2D elliptical geometry is used to restrict the sampling space of the initial molecular distance bounds matrix. We show that this can be used to both control and measure the roundness of the generated conformers.

3. The option for defining custom pairwise force terms in the final conformer optimisation stage was introduced. In this way, a user can add customisable attractive/repulsive interactions to bias conformer sampling based on their knowledge of the macrocycle under study.

The modified conformer generator has been tested on several systems, one of which is a cyclic decapeptide series [3], which adopt a conformation with four intramolecular hydrogens in chloroform as determined by NMR. We are interested to see if the modified conformer generator is able to find this ‘closed’ conformation with intramolecular hydrogen bonds. We find that the modified ETKDG+eccentricity method provides better sampling compared to the standard approach. Sampling can be further improved when selected pairwise interactions are included. A second test system is cyclosporin A, which has a cis-amide bond in the ‘closed’ conformation. We show that our approach allows to force sampling of the desired cis-amide bond while keeping the rest of the amides in the trans configuration. As a third test system, we investigate a diverse set of macrocycles collected by researchers at OpenEye, which includes small macrocycles from both the Protein Data Bank and Cambridge Structural Database. Also for this dataset, the performance of ETKDG+eccentricity was found to be better than the approach.

The improved conformer generator for macrocycles will be made available in the RDKit.

[1] Hopkins A. L. , Groom C. R. The druggable genome, Nat. Rev. Drug Discov., 2002, 1, 727–730.
[2] Riniker, S. Landrum, G. A. Better Informed Distance Geometry: Using What We Know To Improve Conformation Generation, J. Chem. Inf. Comp. Sci. 2015, 55:2562-2574
[3] Fouche, M., Schäfer, M., Blatter, M., Berghausen, J., Desrayaud, S. and Roth, H.J. Pharmacokinetic Studies around the Mono‐and Difunctionalization of a Bioavailable Cyclic Decapeptide Scaffold. ChemMedChem, 2016, 11(10),1060-1068.