Introduction

Macrocycles are cyclic macromolecules containing rings of 8–12 or more atoms [1, 2]. These compounds are found in natural products such as macrolides, that consist of a large macrocyclic lactone ring to which one or more deoxy sugars may be attached, which have antibiotic or antifungal activities and are used as pharmaceutical drugs [3]. Due to their unique properties [4, 5], such as of their ability in achieving protein-protein inhibition [6, 7] and cell permeability [8], this class of compounds has gained an increased interest in drug development. As a consequence, many macrocycles are currently in clinical development [4], supporting the development of macrocycle-based pharmaceutical companies (Cyclenium pharma, Circle pharma).

The knowledge of the 3D structure is fundamental to the rational design of biologically active molecules. Ideally, these structures would be determined experimentally, either by X-ray crystallography or nuclear magnetic resonance spectroscopy, but this could sometimes be a laborious, time-consuming costly avenue (if possible at all). Fortunately, molecular modelling techniques were developed over the past decades to assess the conformational sampling of chemical structures with a reliable accuracy [9]. To our knowledge, the current bioinformatics tools that are available to investigate and predict macrocycles 3D conformations are limited in their availability, precluding the development of macrocycles, as they are commercially distributed (MOE [10], MacroModel [11], Tork [12], Corina [13] or Schrödinger [14]) or not available to the public (SOS [15] and ForceGen 3D [16]).

In this Software Metapaper, we present an open-source tool suite called ConfBuster, providing tools for macrocycles conformational search, analysis and visualisation of results. The tool suite is based on Python and the NetworkX [19] python package, and on Open Babel [17] chemical toolbox and the PyMOL [18] molecular viewer. For the optional analysis of the results, ConfBuster also relies on the R programming language and the ComplexHeatmap [20] and circlize [21] packages. For several examples, ConfBuster found macrocycle conformations that are within few tenths of Å of the experimental structures in minutes. To our knowledge, this is the only open-source tools for macrocycle conformational search available to the scientific community.

Implementation and architecture

ConfBuster is composed of 4 Python scripts with the goal of finding the lowest energy conformation of single macrocycles, given a set of molecular coordinates provided as MOL2 or PDB format. The relation between the ConfBuster scripts is illustrated in Figure 1, along with their respective third-party program and package dependencies. The primary script (ConfBuster-Macrocycle-Linear-Sampling.py) performs a conformational search by cleaving the macrocycle at different positions and, for each of the linear molecule created from the cleavage of a bond, calls the secondary scripts to perform a rotational search (ConfBuster-Rotamer-Search.py) and energy minimisations (ConfBuster-Single-Molecule-Minimization.py). Further, the primary script provides PyMOL input files for visualisation of the search progress and results. As an optional step, the script ConfBuster-Analysis.py generates the plot of the root mean square deviation (RMSD)-based hierarchical clustering of the resulting conformations and of the conformational energies.

Dependencies. ConfBuster depends on Open Babel [17], either for file format conversion as well as several functions for conformational sampling and energy minimisation. ConfBuster also depends on PyMOL [18] for RMSD calculations, for a dihedral sampling protocol and for visualisation. The Python package NetworkX [19] is used to identify atoms belonging to cycles and macrocycles. In addition to standard R packages, the optional ConfBuster analysis relies on the ComplexHeatmap package [20] available through the Bioconductor software project (bioconductor.org) and the circlize package [21]. The respective dependencies of ConfBuster scripts are illustrated in Figure 1.

ConfBuster-Single-Molecule-Minimization.py. This script uses the command-line program obminimize from the Open Babel package to optimize the geometry and minimise the energy of the molecule. Then, the obenergy command-line program is used to calculate the final energy, stored in the title of the final MOL2 file. The script command line arguments are the mandatory input filename (-i name) and the prefix of the output files (-o [default: replace input file]).

ConfBuster-Rotamer-Search.py. This script uses the obabel command-line program with the --conformer option to generate a number of conformations. These conformations are then all evaluated with the obenergy program. Corresponding energy is stored in the title of the resulting molecule coordinate file. The script command line arguments are the mandatory input filename (-i name), the number of generations for the genetic algorithm search (-g [default: 100]), the energy cutoff used to discriminate conformations in units of kcal/mol (-e [default: 50]), the output directory name (-d [default: use the prefix of the input filename] and the format of the outputted molecules (-f [xyz or default: mol2]).

ConfBuster-Macrocycle-Linear-Sampling.py. This is the main script of the tool suite that performs the macrocycle conformational search. The search is achieved by cleaving the macrocycle in linear molecules (Figure 2A). For each cleavable bond (any single bond between two atoms that are not chiral centres), a conformational search is performed. First, the bond is removed and hydrogens are adjusted on the two terminal atoms using PyMOL. This linear molecule is then sampled n times to identify n low energy conformations. For each of these samplings, new conformations are generated from systematic rotations of all the dihedrals angles and all possible pairs of dihedral angles. N clash-free conformations are selected, from the shortest to the longest distance between the cleaved atoms, to be cyclized back and minimized (Figure 2B), resulting in a total of n*N cyclized conformations per cleavable bond. The script command line arguments are the mandatory input filename (-i name), the root mean square deviation cutoff in units of Å (-r [default: 0.5]), the number of rotamer searches for each cleavable bond (-n [default: 5]), the number of conformations kept from each rotamer search (-N [default: 5]) and the mandatory output directory name (-o name [default: prefix of the input filename]).

ConfBuster-Analysis.py. In this optional step, the RMSD values between all the conformations are calculated. The RMSD- and energy-based classifications are listed in text files and the hierarchical clustering of the conformations based on the Euclidean distances between the RMSD values is plotted in a tree and a 2D matrix, and the energy of each conformation is plotted using a colour scale (see Figure 2D). This publication-quality figure, stored in a PDF file, allows the quick identification of the lowest energy conformation as well as an RMSD clustering of the best-energy conformations from the search. The script command line arguments are the mandatory directory name of the search results (-i name), the root mean square deviation cutoff in units of Å (-r [default: 0.5]), the number of conformations to include in the analysis (-n [default: use all the conformations present in the directory]) and the mid-point value of the energy color scale in units of kcal/mol (-e [default: 0]).

Results. Figure 2B and 2D presents the search results for the macrocycle sopharen A from PDB 1W96 [22]. Figure 2B exposes the 73 conformations identified from the search, Figure 2C presents the lowest energy conformation aligned to the reference structure from the PDB file (the RMSD between the two conformations is 0.405 Å), while Figure 2D displays the RMSD hierarchical clustering and the energy-based classification of the conformations identified from the search. This last figure allows the identification of the lowest energy conformation as well as to evaluate the relation between the conformations. Several examples of a molecular conformational search are provided with the ConfBuster distribution. In these examples, the RMSD values between the best search results and their respective reference conformations range between 0.010 Å (PDB 3R92 [23]) and 2.728 Å (PDB 3MT6 [24]).

Quality control

ConfBuster installation, dependencies and running instructions are provided in details with the distribution on the Github repository. Additionally, several examples including all the command lines and required files to run macrocycle conformational searches are also included with the distribution, in the examples folder and in the examples/Instructions.pdf file. As for the sopharen A example above, the macrocycle molecule was extracted from the PDB 1W96, and the molecule was validated for the correct bond orders and the hydrogens were added and saved in the file examples/1w96/macro-1w96.pdb. Then, a minimisation was performed, using the following command in a terminal window:

$ ConfBuster-Single-Molecule-Minimization.py -i macro-1w96.pdb

This was followed by a macrocyclic conformational search:

$ ConfBuster-Macrocycle-Linear-Sampling.py -i macro-1w96.mol2 -n 5 -N 5 -r 0.5

The progress of the search may be monitored using the run command in PyMOL:

(in PyMOL) run Follow-macro-1w96.py

Finally, the analysis of the search results was performed as follows:

$ ConfBuster-Analysis.py -i macro-1w96 -R macro-1w96.mol2 -n 20

which built the Heatmap_20.pdf file with the content presented in Figure 2D.

Therefore, the users can assess that their installation is working properly by cross-checking their results against those presents in the distribution. Further, the users can also validate the results obtained using other molecules against experimental results whenever available. However, as the searches involve random parameters and different molecular complexity, the results might be slightly different than those included in the code distribution or from the experimental values. Running multiple conformational searches may increase the liability of the results. In any case, support is available and provided via GitHub Issues.

Operating system

ConfBuster is able to function on any operating system that supports standard Python and R installations and the dependent packages, which includes Linux, Windows and macOS (tested on Linux Ubuntu 14.04 LTS).

Programming languages

Python == 2.7 and (optional) R ≥ 3.0.0 (tested with Python 2.7.6 and R 3.4.1).

Additional system requirements

There is no special system requirement. However, hardware requirements in terms of processor power, memory capacity, etc. depend primarily on the complexity of the molecules that are processed.

Dependencies

The following software is a required dependency for all the ConfBuster scripts:

Open Babel == 2.4.1 (will not work with older or more recent versions, tested with Open Babel version 2.4.1).

The following Python package is a required dependency for ConfBuster-Macrocycle-Linear-Sampling.py:

NetworkX (tested with version 1.11).

The following software is a required dependency for ConfBuster-Macrocycle-Linear-Sampling.py and ConfBuster-Analysis.py, and optional for the other ConfBuster scripts:

PyMOL ≥ 1.8 (tested with version 1.8.5.0).

The following R package is a required dependency for ConfBuster-Analysis.py:

ComplexHeatmap (tested with ComplexHeatmap version 1.14.0, from Bioconductor release 3.5 (bioconductor.org)).

Circlize (tested with version 0.3.10).

List of contributors

1. Xavier Barbeau: Development.
2. Antony T. Vincent: Contributions, documentation and testing and code review.
3. Patrick Lagüe: Documentation and testing.

Software location

Language

English