phenix_logo
Python-based Hierarchical ENvironment for Integrated Xtallography
Documentation Home
 

Overview of molecular replacement in PHENIX

Input files and mandatory parameters
Outline of MR procedure
Limitations
Frequently Asked Questions
Reference

Molecular replacement (MR) is a phasing method that uses prior information in the form of a related or homologous structure. The procedure is roughly divided into two steps, a rotation function (RF) to determine the orientation of the search model(s), and a translation function (TF) to determine the absolute position(s) in the unit cell. Because it requires no additional experimental procedures or data, and additionally simplifies model-building, MR is usually the method of choice for structure determination when a suitable search model is available. (See the Limitations section below for advice on search models.)

In PHENIX, MR is performed by the program Phaser, written by Randy Read's group at the University of Cambridge. Although Phaser may be run on the command line with CCP4-style inputs, we recommend using either the AutoMR wizard (GUI or command line), or the Phaser-MR GUI. AutoMR presents a user-friendly and automated frontend to Phaser, and interfaces directly with the AutoBuild wizard for model-building. We recommend that new users, and anyone who expects MR to work for their specific structure, start with the AutoMR GUI. The Phaser-MR GUI is more complex, but enables finer control over parameters and multi-step searches, which may be necessary for difficult structures. The Sculptor and Ensembler utilities are available for preparing search models.

(Phaser is also used for experimental phasing, but this functionality is exposed through the AutoSol wizard and Phaser-EP GUI.)

For new users, we recommend reading this document first, followed by the Phaser-MR tutorial for a step-by-step introduction to running MR in the Phenix GUI.

Input files and mandatory parameters

All of the MR programs in PHENIX require a single reflections file containing experimental data (with sigmas); AutoMR and the GUI will accept any file format or data type, including intensities. The procedure traditionally uses all reflections, so R-free flags are not required. (Unlike refinement, this does not severely bias the final R-free value, since the placement of molecules will be approximate and some conformational changes usually occur.)

At least one search model is required. In most cases this will be a PDB file containing a partial structure. For more variable structures, an ensemble model may be used instead - either a PDB file with multiple MODEL records, or multiple similar PDB files. When using an ensemble, all models must be superposed in the same orientation; the Ensembler utility is used for this. There are no limitations on the size or number of search models. For large complexes, if the relative positions of individual subunits do not change, the entire assembly may be used as a search model instead of placing each component separately. All heteroatoms (ligands and waters) should be removed from the PDB file(s) before use.

Another option, only available in Phaser itself (not AutoMR), is to search using a map (or rather, an MTZ file containing pre-weighted map coefficients), often solved at low resolution. This requires additional information about the center and extent of the map section to search with.

The maximum likelihood phasing methods used in Phaser require prior knowledge about the deviation (or variance) of the search model(s) from the real structure, and the expected composition or scattering mass of the crystal. To specify the model variance, either an RMSD value or percent sequence identity may be used (these will be converted internally, and a sequence identity of 100% is assumed to mean an approximate RMSD of 1.0A). It is important to minimize the variance if possible (see Limitations below for guidelines), which often requires eliminating atoms or modifying B-factors. The Sculptor utility will perform this step, given a PDB file and a sequence alignment. (This is usually unnecessary for search models with high sequence identity to the target molecule.)

For composition, you may supply a sequence file (protein or nucleic acid), or simply enter the molecular weight. The standalone versions of Phaser also accept the fractional composition of each search ensemble, if known. Note that the composition data does not necessarily have a 1:1 correspondence with the search ensembles (see FAQ list below for details). Even if you are only searching for a single ensemble out of several (e.g. the protein in a protein-DNA complex), you must still supply the expected composition of the entire crystal.

Outline of MR procedure

The automated molecular replacement method in Phaser involves several discrete steps:

  • Anisotropy correction: scales reflections as necessary to overcome anisotropy (weak data in a particular direction).
  • Rotation function: identifies orientation of model(s).
  • Translation function: given the orientation(s) from the RF, finds the absolute position(s) in the unit cell.
  • Packing analysis: filters TF results based on number of clashes between atoms, given a certain cutoff. (By default, this uses a very conservative cutoff, which may not be ideal for searching with distant homologues.)
  • Refinement and phasing: performs simple rigid-body refinement of the placed molecules, and calculates phases from the final solution.
  • Log-likelihood gain calculation: determines the final LLG, which can be used to evaluate the success of MR.

If multiple search models are used, these steps will be performed sequentially for each model. Although each step may be run individually in the Phaser-MR GUI, this is necessary only in exceptionally difficult cases.

Limitations

The main restriction on the use of molecular replacement is the requirement for a suitably similar search model. Although there is no exact rule for this, the relationship between sequence identity and MR success is roughly as follows:

  • Better than 40% identity: usually easy, unless large conformational changes are involved.
  • 30-40%: MR possible, but sometimes more difficult.
  • 20-30%: MR difficult, careful model preparation required.
  • Less than 20%: MR unlikely to work in most cases.
  • In terms of RMSD, above 2.5A is very unlikely to work, while 1.5A or less is preferrable.

Structures which undergo large conformational changes may need to be split into separate domains for searching, regardless of sequence identity. Where multiple similar search models are available, combining these into an ensemble may improve the likelihood of success. Processing models with the Sculptor utility is highly recommended, especially at lower sequence identity.

Additional problems include:

  • Low resolution; although MR often works best with a reduced set of reflections, the assumption is that these are measured very accurately, which is not the case when the overall resolution limit is already 4A or worse.
  • Packing clashes due to model deviations and/or extra residues. This can result in an otherwise valid solution being thrown out. Removing the offending residues or lowering the packing cutoff can circumvent the problem.

For cases where anomalous data from a SAD experiment are available, a poor (but genuine) MR solution may be used to identify heavy-atom sites and combined with SAD phases, a technique known as MR-SAD. This may provide a decent-quality map where neither technique is independently sufficient. The AutoMR and AutoSol manuals have details on running this in Phenix.

Frequently Asked Questions

The Phaser home page has a general FAQ list; look there first to see if it answers your question. Note that most of the questions below apply mainly to the "automated molecular replacement" (MR_AUTO) mode of Phaser.

  • How do I know if Phaser has solved my structure? This is answered in detail in the section "Has Phaser solved it?" in the main Phaser manual, but it can be summarized as follows: the final translation function Z score (TFZ) should be above 8, and the log-likelihood gain (LLG) should be positive and as high as possible. Of course, as with experimental phasing, the ability to autobuild most of the model is the single best measure of a correct solution; alternately, running a single round of refinement should result in an R-free below 0.50.
  • Why doesn't Phaser report an R-factor? The R-factor is far less sensitive than the TFZ and LLG scores, especially when searching for remote homologs or a small part of the overall scattering mass. Phaser does record R-factors for the rigid-body refinement step towards the end of a run, which can be found in the logfile. However, unless these values are relatively low (< 40%), they are not reliable indicators of solution quality, which is why only the Z-scores and LLG are reported in the GUI.
  • What if I don't get a good solution? This is also covered in detail in the main Phaser manual, in the section "What to do in difficult cases".
  • What resolution cutoff should I use? By default Phaser uses all data to 2.5 A. We recommend leaving the resolution alone, but in some cases reducing it may improve results (it will also lead to a significant improvement in speed). The effects of bulk solvent at low resolution are already taken into account internally, so you do not need to specify a low-resolution cutoff either.
  • Should I use the output MTZ file from Phaser for refinement? No, the amplitudes (Fs) in this file have been corrected for anisotropy, which is helpful for the MR search but may significantly alter them from the experimental values. The MTZ file is only useful for map coefficients, but in most cases you will find that much better maps can be obtained by immediately running a round of refinement or automated re-building.
  • How are the composition and search ensembles related? Technically, they aren't connected. The ensembles are used in the actual MR search; the composition only tells Phaser what to expect for the overall scattering mass of the asymmetric unit (an important factor in the maximum likelihood scoring). Although you may only specify one sequence per component, you can instead provide a molecular weight for a collection of chains. You may also split a model into smaller ensembles (for instance, different domains) while keeping the composition the same, or use a multi-chain search model while specifying a sequence file for each chain.
  • How can I make Phaser go faster? The new MR_FAST mode will reduce the redundancy of multi-copy TF searches, and can significantly reduce overall runtime. In many cases, adjusting the resolution limit will dramatically increase speed, at some cost to accuracy (depending largely on model similarity).

See the Phaser-MR GUI manual for additional details specific to that interface.

Reference