4.9. 3D Conformers
Starting in ADMET Predictor version 10.4, users can generate 3D conformers for one or more input molecules, both within the graphical interface and from the command line. For the graphical interface, the new 3D functionality is available under the menu item TOOLS >> 3D, as well as from the main toolbar.
Command-line functionality is available through script-file workflows such as ms_generate_3d.txt, ms_mmff_minimize.txt and ms_align3d.txt, as described in the Script File Workflows section.
Compounds in the spreadsheet are always represented using 2D depictions, but each can have a single 3D conformer stored internally. The Geometry spreadsheet column will indicate "3D" if such a conformer is stored for a given compound. 3D conformers, whether read from file or generated within the ADMET Predictor interface, are retained during the session and are saved automatically in XTK files along with the 2D structures. They can be exported to file in SDF format and used as inputs for predicting the subset of ADMET Predictor properties that are based on 3D descriptors, as described in the Calculating ADMET Properties section.
To visualize the 3D structure of a single compound, you can use the menu item TOOLS >> 3D >> Launch 3D Viewer or the corresponding toolbar button. This feature requires that you first specify an external viewer application on the 3D tab of the main Preferences window.
It is important to note that once a chemical structure has been modified, such as by adding or removing atoms, any previously generated 3D structure will no longer be valid. It is recommended that 3D structures be regenerated after any such modification.
Methodology
By default ADMET Predictor generates a single 3D conformer for each input molecule. A brief overview of the methodology is provided below, with the subsequent section describing the option to generate multiple conformers. This and other options are described in the Preferences section.
The first step is to partition the molecule into its individual ring systems, which are processed independently. Two different algorithms are available for generating 3D ring-system coordinates, each of which yields a small number of diverse candidate conformers. The first is based on assembling individual ring conformations derived from predefined templates, and includes a backtracking protocol to avoid ring fusions that result in collisions with the growing system. A second processes the entire ring system at once using a variant of distance geometry, with multiple random embeddings used to generate a set of candidate conformers.
By default the ring-assembly method is used for all ring systems classified as simple: namely, those that contain no bridged fusions and no rings with more than seven atoms. All other ring systems are handled using the distance geometry algorithm, though there is an option to use this algorithm for all ring systems, as discussed below in the section on multiple conformers. Regardless of the method used, each candidate ring-system conformer undergoes a final minimization using the MMFF94s force field (Halgren 1996), and the conformers are ordered by increasing energy.
After all ring systems have been processed, coordinate generation for the molecule enters an initial phase whose goal is to produce a reasonable, conflict-free conformer quickly. It employs a breadth-first search (BFS) of the molecule, assigning coordinates to atoms one at a time. When a ring atom is encountered, the predefined conformation of the entire ring system is positioned relative to the growing structure. Bond lengths are based on the MMFF94s force field, while bond angles are assigned based on atom hybridization states. The three additional degrees of freedom are the potentially multiple conformations of each ring system that were generated in a previous step; the two mirror-image substituent arrangements of tetrahedral atoms that do not have an assigned stereo configuration; and torsion angles, which are chosen from a fragment-based preference table derived from literature rules and publicly available structures. These three degrees of freedom are explored locally when each atom is positioned during the initial BFS, with the goal of avoiding overlap with the portion of the molecule that has already been considered.
Once the initial BFS over the molecule is complete, a second optimization phase is used if any atoms are still in close contact. This phase involves a more thorough exploration of the available torsion space, and it terminates as soon as a conflict-free conformation is discovered.
The final step is to perform a minimization of the single 3D conformation. This is performed only if the relevant option is specified, as described in the Preferences section.
Generating Multiple Conformers
Users can choose to generate multiple conformers for each input molecule. The protocol begins with the single-conformer algorithm described earlier, but in this case multiple initial conformers are saved comprising different conformations of the individual ring systems. For each of these initial conformers, multiple new conformers are generated using an exhaustive exploration of torsion space, with the size of the space determined by the Conformer Search Depth setting described in the Preferences section.
The final collection of conformers is then sorted by increasing energy, and a user-specified number of lowest-energy conformers is selected, ensuring that the total energy range and RMSD between conformers is within user-specified limits. These and other user options are described in the Preferences section.
Note that in order to increase the number of ring-system conformers, it may be useful to turn on the setting to use distance geometry for all ring systems. See the Preferences section for a description of this option.
Ring System Templates
It is sometimes useful to constrain a portion of a molecule to have a specific, predefined 3D geometry. ADMET Predictor has such a feature, though it is currently restricted to complete ring systems, defined as the set of ring atoms/bonds to which no non-ring bonds are connected.
To use this feature, you must first create a 3D template file defining the desired 3D geometry. The script-file workflow ms_create_3dtemplates.txt is used for this purpose. It uses as input a 3D SDF file containing the 3D geometry of interest and produces as output a text file with a description of the motif and its 3D coordinates. Please see the notes in ms_create_3dtemplates.txt for further details.
Once you have created the template file, it can be used for 3D conformer generation either in the graphical interface or from the command line. In the former case, the path to the file is specified in the main Preferences window. In the latter, it is specified in the script file ms_generate3d.txt.
A second option controls whether or not the predefined template should undergo minimization during coordinate generation. This can be useful for resolving any conflicts that arise when defining coordinates for the ring-system substituents. However, it is possible that the minimization could significantly alter the ring-system coordinates, so users are encouraged to experiment with this setting.