ConfGen is a new program designed to perform rapid conformational searches that result in a close approximation of a ligand's bioactive geometry. Here, ConfGen product manager Dr. John Shelley discusses the motivation for and implementation of the program.
Overview
While the ability to generate diverse conformations for drug-like molecules is of central importance to drug design research, it is often necessary for researchers to strike a suitable compromise between how much time and computer resources are used versus the accuracy of the results. Though it may not always be clear where the best balance between rigor and practicality lies, it is, broadly speaking, desirable for conformational search algorithms to generate as close an approximation of a ligand's bioactive conformer as possible. However, certain real-world applications demand that completeness of conformational sampling should not come at the expense of practicality, either for the conformational search itself or for downstream calculations on the conformer set. The results of internal research at Schrödinger into how best to arrive at bioactive conformers within the smallest conformer set possible led to the creation of ConfGen as a standalone program.
ConfGen is a program designed to efficiently and accurately generate ligand conformations. A wide variety of conformational search protocols are supported in a convenient manner, giving researchers the ability to choose the approach that best suits their needs. ConfGen excels at producing bioactive conformers within a small number of total conformations: in our validation studies described, below, ConfGen has a 96% success rate reproducing the bioactive conformation within 2 angstroms, while producing an average conformation set of just 16.3 conformers per ligand. This efficiency improves the results from and reduces the time spent in downstream applications such as pharmacophore searching, shape-based screening, or flexible superposition.
ConfGen Methodology
For a number of years, ConfGen technology has been the core of the conformational search engine in Glide, Schrödinger's ligand-docking program. More recently, a variant has been available as the Ligand Torsion Search module in MacroModel. With significant enhancements and validated search protocols, this capability is now a standalone program. ConfGen's three main themes for introducing conformational variation are:
- Variation of rotatable bonds
- Rotatable bonds in linear portions of the ligand are identified
- Minima for each rotatable bond are located and ranked using a simplified force field
- Variation of flexible ring systems
- Flexible ring systems are identified
- Conformational templates are sought from a library of almost 1200 ring systems
- Inversion of sp3 N atoms bonded to 3 other atoms
Above,
an example of ConfGen's systematic conformer generation: Exploring
rotatable acyclic bonds (a), exploring flexible rings (b), and the
resulting set of initial conformers (c).
Up to a total of 20,000 initial distinct conformations are generated for each ligand by systematically sampling all combinations of minima for rotatable bonds1, ring system conformations, and N atom inversions (including symmetry). These initial structures are rapidly filtered to eliminate redundant or energetically inaccessible conformations. The exact filtering criteria used depend on which options the user selects. The standard ConfGen interface distills a variety of options down to "Fast," "Very Fast," "Intermediate" and "Comprehensive" protocols, with a broad range of adjustable settings revealed to those who wish to use ConfGen's advanced interface. These four basic ConfGen protocols have been the subject of internal validation studies, and use the following criteria to filter crude conformers:
- Very Fast (no energy filtering) - Generates up to 5 conformers per degree of freedom. As conformers are processed, eliminates conformers if they have an RMSD value of less than 1.25 Å and all of the dihedral angles involving polar hydrogen atoms are within 60 degrees relative to a previously accepted conformer.
- Fast - Similar to Very Fast, except that an RMSD value of 1.0 Å is used to detect redundant conformers, and conformers whose energy is more than 25 kcal/mol (104.67 kJ/mol) higher than the lowest energy conformer are eliminated.
- Intermediate - Similar to Fast, except that 75 conformers are generated per degree of freedom and the quality of ConfGen sampling is enhanced.
- Comprehensive - Similar to Intermediate, except that an RMS value of 0.5 Å is used to detect redundant conformations and the energy window for eliminating candidate structures is increased to roughly 120 kcal/mol (500 kJ/mol).
Additionally, the basic ConfGen interface allows users to minimize input and/or output structures when using the Fast, Intermediate, or Comprehensive mode.
ConfGen validation results
The performance of a conformational searching method can be measured by its ability to reproduce the bioactive conformations of ligands from experimentally determined structures of protein-ligand complexes. The table below gives these results for ConfGen for a diverse set of more than 650 ligands.
| Method | Percent bioactive conformations matched within the distance given | Average Number of Conformers | Average CPU time (s/ligand)2 | |
| < 1.0 Å | < 2.0 Å | |||
| Very Fast | 54% | 96% | 16.3 | 1.0 |
| Fast | 55% | 95% | 13.5 | 2.3 |
| Intermediate | 64% | 96% | 38.8 | 6.5 |
| Comprehensive | 71% | 96% | 153 | 16.3 |
We believe that ConfGen results match or exceed the quality of those produced by competing programs. By design ConfGen generates structures resembling low energy conformers so minimization improves reproduction of bioactive conformations only slightly. Since minimization increases the CPU time by a factor of 4-12, for studies that focus on shape as opposed to energy, eliminating minimization is an attractive option. Finally, the faster methods produce results that are nearly as good as the Comprehensive approach suggesting that in many circumstances, they are viable approaches for performing studies involving many ligands.
ConfGen Features
ConfGen leverages the versatile molecular modeling program MacroModel, providing a wide range of functionalities:
- Minimize input and output structures employing a range of force fields and solvation treatments
- Redundant conformer elimination and energy filtering capabilities
- Conformer subset selection
- Parallelization across multiple processors
ConfGen also has these additional features:
- New panels supporting validated search approaches accessible via intuitive controls
- Very fast, Fast, Intermediate, and Comprehensive search protocols
- Optional minimization of structures
- An Advanced panel permits the user to vary many aspects of ConfGen searches
- Better sampling of weak torsional potentials
- Avoiding conformations with very unfavorable electrostatics
- Elimination of conformations with high levels of intramolecular contacts
- Variable sampling based upon the effective number of degrees of freedom
- Takes into account the number of rotatable bonds, ring conformations and nitrogen atom inversions
1 Terminal rotatable bonds often have a smaller impact on the overall shape of a molecule. Depending on the search and options selected, ConfGen will either sample all combinations of the minima for such bonds, or place all of them in their lowest energy minima and generate conformations by sampling higher energy minima one terminal bond at a time..
2 Times for a single core on a 2.2 GHz dual-core Opteron 275