Evaluation Planning Guide
- Know the method: What problems does it solve?
- Know the method: What metrics should be used?
- Steps to a successful evaluation
- Hardware and software dependencies
- Licensing: Want to give it a try?
- More information about the method
Through numerious evaluation efforts in recent months, we have learned several lessons that should be applicable to users at your site as well:
Know the method: What problems does it solve?
Phenix/DivCon is primarily a structure or energy minimization technology that applies an advanced molecular modeling, quantum mechanics (QM) functional to the X-ray refinement process. It is therefore applicable to the following problems:
General structure refinement of protein, DNA, RNA, fats, and ligand(s): With an advanced QM functional, most biologically relevant chemical systems can be treated without any special configuration or parameters.
Binding mode refinement and characterization: QM refinement is reliant on the initial structure and the capabilities of the optimizer within Phenix, but if you pair the method with a density-aware docker or with intuition, you can explore the active site and use the metrics listed below to consider the final binding modes and choose the most logical one. You will find that each binding mode is chemically correct vis–à–vis bond lengths, angles, and so on, but certain modes will find the right “fit” with the target.
Protonation and tautomer exploration: The QM method is exquisitely sensitive to the chemistry of the structure including not only bond lengths, angles, and torsions, but also to non-bonding interactions like electrostatics, polarization, hydrogen bonding, dispersion, and so on. To use this method, run multiple, parallel QM refinements each with different protonation states, and use the metrics below to decide which is correct. The recently published XModeScore tool is provided for this purpose!
“Exotic” chemistry refinement: With support for 70 elements, the PM6 Hamiltonian implemented in DivCon is able to support metal containing systems, unusual ions, covalently bonded ligands, and so on – usually with very little if any special settings or parameters. The only requirement at this time is that the structure have an even number of electrons (or is closed shell).
What problem is the method not designed to solve? It does not include any pre-refinement docking or sampling, so the method is designed to explore a local potential energy surface. Therefore, the concept of garbage-in/garbage-out is important to keep in mind. If you have controversial protonation, tautomerization, or binding modes to explore, you should run parallel refinements from different starting states and use the metrics or indicators listed below to help choose the correct final state(s). The benefit of the more advanced QM functional is that it is generally more trustworthy then conventional stereochemical restraints, and when it is joined with experimental density – such as in QM-based refinement – the two “sides” come together in a complementary fashion.
Know the method: What metrics should be used?
In our experience, there are four crucial metrics or indicators that should be considered when exploring the method for use in your drug discovery or design workflow:
Agreement with experimental density: While the density is used within the refinement, so is the calculated gradients from the quantum mechanics functional. We have therefore observed that poor protonation or poor starting ligand geometry (initial binding mode) can lead to positive/negative density within the active site upon completion of the refinement. This is not a bug – protons are important! Often, “hot” ligand conformation energies are probably indicative of an incorrect pose, and they should be thrown out accordingly. With access to a more advanced functional, you can use the data provided to make actionable decisions!
Ligand strain: As has been consistently shown, the strain on the ligand is a good indicator of poor chemical accuracy in the ligand. This relationship carries through even to rigorous ab initio calculations. The pre-refinement and post-refinement ligand strains are reported within the phenix.log. Simply search the phenix.log file for STRAIN and both values will be reported for all chosen ligands.
Primary Interactions: Using the Molecular Operating Environment (MOE) from CCG, you can report the basic or primary interactions that drive binding. These include hydrogen bonds, arene-bonds, ionic bonds, covalent bonds, and so on. Tools are provided with the package to help capture this information and graphically report it for use in comparisons between refined structures. The benefit of these “primary” interactions is they are often less controversial then chosen levels of theory for interaction (or binding) calculations using quantum mechanics and molecular mechanics.
Interaction Energy Calculation: The DivCon Discovery Suite is able to calculate the interaction energy (Eint), along with the decomposed terms of Epol, ECT, and Ees. The EINT performed on the target:ligand complex is based on the chosen Hamiltonian and is quite accurate and can be used to compare different poses and other chemical relationships.
XModeScore: The recently published XModeScore method is provided for this purpose.
Steps to a successful evaluation
With the strengths and abilities of the method in mind, what are some steps to a successful evaluation? Granted, these steps can be performed in any order, but these are the steps it seems that most people take in their exploration of the method. To understand the method and its ability to shorten the conventional “refinement/tweak/re-refinement cycle,” it is important to move as quickly as possible beyond retrospective re-refinement of old structures and move to current, “real life” examples:
- Tutorials: Start with the provided tutorials and make sure that the method has been installed and configured correctly. It is also good to review the FAQ and the Guidelines to get a better feeling for the experiences of our users.
- Retrospective Refinement: The first stage of most evaluations is to rerun several old or retrospective refinements and see what changes. This step is important to validate the software; however, this step should be seen as a beginning – not an end – because it does not demonstrate the power and flexibility of the method. Phenix/DivCon is not doing any sampling so you will end up with lower energy versions of the structure you started with. However, you will also probably rescue interactions, and these are important changes when it comes to vHTS, structure-based ligand design, etcetera. For a crystallography evaluation, the next steps are much more interesting.
- Prospective Refinement: Next, once you trust that the refinement is working through retrospective refinement, try prospective refinement. With retrospective refinement, it is easy to discount the amount of work that goes into refining the structure the first time. Prospective refinement allows you to experience how much easier it is to get to a more reasonable geometry. Remember, since Phenix/DivCon does not utilize CIF files and other stereochemical restraints or link terms, any chemical interactions that are formed are based upon quantum mechanics – not a priori assumptions or biases.
- Redocking/Replacement: On that vein, move on to running parallel refinements on several different starting (docked) poses or protonation states and compare the metrics listed above to “choose” the right final binding mode or protonation state. Initial conditions are so important in refinement, and therefore, each of the docked poses should end with different final binding modes. In other words, use docking to explore the space, and our tool to help refine/choose the right binding mode!
- Protonation or Tautomeric States: Since the method is sensitive to protonation, you can use the method to explore controversial protonation states. To do so, run several, parallel refinements – each with different protonation states – and compare the final structures using the above noted metrics to choose the correct protonation state(s). The recently published XModeScore tool is provided for this purpose!
- Exotic Chemistry: So far, none of these steps were really focused on unusual chemistry. Are there projects involving metal-containing systems or other exotic ionic situations? PM6 supports 70 elements. What about covalently linked ligands, macrocycles and other flexible molecules, protein:protein structures, and so on? Once you have reached this point in the evaluation, you will have become an expert in the method, and you can really stretch the method – potentially beyond the points currently explored by QuantumBio staff.
Hardware and Software Dependencies
Generally, for most types of calculations, QuantumBio’s software has minimal dependencies (especially next to other quantum mechanics software). These are generally broken down into the following items:
Processor: Any 64bit, Intel x86 compatible processor will do. The more processor “cores” available to the operating system the better as the DivCon Discovery Suite is built for multi-threaded, multi-core processing.
Memory: As a general rule, memory grows both with the number of QM-atoms/residues (e.g. core+main+buffer) and the number of processor cores. Often 1GB or 2GB per processor core is enough to run most calculations, and refinements can be routinely executed on a laptop.
Operating System: Currently, Linux64 and Mac OS X 10.7.5+ are the primary operating systems supported by QuantumBio. The Linux64 version is built on RedHat Enterprise v.5.x so any operating system compatible with that API/ABI should work with the executable provided. We have clients who use the Suite on CentOS v.5-7, Debian, RedHat Enterprise v.5-7, Amazon, and several other distributions. If you have a particular distribution in mind, please feel free contact us for details about this support.
Third Party Software: In order to make use of the package, some third party software software is required:
- Refinement Suite: The Phenix/DivCon plugin relies on the availability of the Phenix platform. Note: if you are interested in the method but do not have Phenix, please Contact Us to discuss alternate options.
- Visualization: Most modern visualization packages will be able to provide the proper visualization capabilities including MOE from CCG, Maestro from Schrodinger, PyMol, Coot, and others. Visualization and setup “plugins” are provided by MOE (written in the native, SVL language provided by the platform).
- Structure preparation: Generally, PDB structures as provided by the RCSB or by corporate databases are “rough” by the standards required by “physics based” methods. Therefore, some structure preparation is required including protonation. Note: for XModeScore, currently MOE is required for protonation and XModeScore will not work without MOE/batch being available on the command line. Coming Soon: the DivCon Discovery Suite will include its own protonation tool which will be used in place for MOE, Phenix ReadySet!, and so on. If you are interested in the method but do not have MOE, please Contact Us for an advanced copy or an update on availability!
Do you have limited time to try it yourself? Try our available Services.
- Guidelines: X-ray Refinement
- Phenix/DivCon: Frequently Asked Questions
- Phenix/DivCon Usage & Tutorial