General

How should I cite PHENIX?

If you use PHENIX please cite:

Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. D. Liebschner, P. V. Afonine, M. L. Baker, G. Bunkóczi, V. B. Chen, T. I. Croll, B. Hintze, L.-W. Hung, S. Jain, A. J. McCoy, N. W. Moriarty, R. D. Oeffner, B. K. Poon, M. G. Prisant, R. J. Read, J. S. Richardson, D. C. Richardson, M. D. Sammito, O. V. Sobolev, D. H. Stockwell, T. C. Terwilliger, A. G. Urzhumtsev, L. L. Videau, C. J. Williams and P. D. Adams. Acta Cryst. (2019). D75, 861-877

Individual programs may have additional citations.

What kind of hardware do I need to run Phenix?

A set of loose guidelines is here. To summarize: most consumer-grade systems (including laptops) purchased in the last three years should be sufficient, and at least 2GB of RAM per processor core is recommended. In many cases older hardware is fine and we ourselves continue to run Phenix on (relatively high-end) systems from 2008, but having lots of memory is still essential.

How can I use multiple processors to run a job?

Only Autosol, MR-Rosetta, Autobuild, Ligandfit, phaser.MRage, phenix.refine, the structure comparison GUI, and phenix.find_tls_groups support runtime configuration of parallel processing. In most cases this is done by adding the "nproc" keyword, for instance:

phenix.autobuild data.mtz model.pdb seq.dat nproc=5

Equivalent controls are usually displayed in the GUI. Some of these programs (phenix.refine is the major exception) also support parallelization over queuing systems; this may require manual input of the queue submission command.

In addition to these options, it is also possible to compile phenix.refine and Phaser with the OpenMP library, which automatically parallelizes specific instructions such as the FFT. This requires using the source installer for Phenix, and adding the argument "--openmp" to the install command. Because of threading conflicts, OpenMP is not compatible with the Phenix GUI.

How can I run Phenix across multiple computers in our lab?

You will need to install a managed queuing system to handle job submission. Currently we support (to varying degrees) Sun Grid Engine, PBS, LSF, and Condor. Note that this assumes that the available processors are dedicated for cluster use, otherwise the submitted jobs will compete for resources with local processes.

You can use the queuing system in two ways: either to parallelize the execution of subprocesses for various programs listed above, or to submit jobs from the Phenix GUI. However, these two options are not inherently compatible: if you submit a job from the GUI it will usually be unable to submit additional child processes to the queue without further configuration.

Where can I find sample data?

You can find sample data in the directories located in $PHENIX/examples. Additionally there is sample MR data in $PHENIX/phaser/tutorial.

In the Phenix GUI, tutorials may automatically be set up as projects. (You must run the specific programs manually, however.)

Can I easily run autosol, autobuild or ligandfit with some sample data?

You can run sample data with with a simple command. To run p9-sad sample data with autosol, you type:

phenix.run_example  p9-sad

This command copies the $PHENIX/examples/p9-sad directory to your working directory and executes the commands in the file run.sh

What sample data are available to run automatically?

You can see which sample data are set up to run automatically by typing:

phenix.run_example  --help

This command lists all the directories in $PHENIX/examples/ that have a command file run.sh ready to use. For example:

phenix.run_example  --help

PHENIX run_example script. Fri Jul  6 12:07:08 MDT 2007

Use: phenix.run_example example_name [--all] [--overwrite]
Data will be copied from PHENIX examples into subdirectories
of this working directory
If --all is set then all examples will be run (takes a long time!)
If --overwrite is set then the script will overwrite subdirectories

List of available examples:  1J4R-ligand a2u-globulin-mr gene-5-mad
p9-build p9-sad

Are any of the sample datasets annotated?

The PHENIX tutorials listed on the documentation index will walk you through sample datasets, telling you what to look for in the output files. For example, the Tutorial 1: Solving a structure using SAD data tutorial uses the p9-sad dataset as example. It tells you how to run this example data in autosol and how to interpret the results.

How is non-crystallographic symmetry handled in PHENIX?

The treatment of NCS depends on the specific program or task:

• In density modification (performed internally by RESOLVE, and used in Autosol, autobuild, phenix.multi_crystal_average, and related programs), NCS relationships (in the form of rotation/translation matrices) are used to average different parts of the map together, which usually improves phase quality. The NCS operators can be derived from heavy-atom positions, the map itself, or a partial model, but this is normally done automatically (although we recommend specifying the number of NCS-related copies if you know what it should be).
• In model-building (Autobuild), NCS is also used to copy pieces of the partial model to complete related chains.
• In refinement (both in phenix.refine itself, and in the Autosol/Autobuild pipelines), NCS-related chains will be restrained to have similar conformations. By default this will be done in torsion-angle space, which allows for local conformational changes if present, and also incorporates advanced sidechain-fitting features. The NCS groups will be determined automatically based on sequence comparison. Alternately, you may restrain the NCS groups to be globally similar, treating the relationships as rigid. This may work better in a few cases but is generally not recommended, especially since careful manual setup is usually required for optimal results.

Note that several programs may produce multiple forms of output specifying the NCS relationships. Files ending in .ncs_spec contain rotation and translation matrices for use in RESOLVE. Files ending in .phil contain atom selections for use in phenix.refine. In most cases, however, you can ignore these files, as the NCS relationships will be detected automatically.

How can I get Phenix to use a local PDB mirror instead of fetching entries remotely?

The Phenix GUI, phenix.cif_as_mtz, MRage, MR-Rosetta and several other programs use a common central function to download PDB files. You can substitute your own local PDB mirror by defining environment variables, for instance in bash/sh:

export PDB_MIRROR_PDB=/data/pdb_mirror/pdb
export PDB_MIRROR_STRUCTURE_FACTORS=/data/pdb_mirror/structure_factors

The structure factors are optional and may be left out to conserve disk space; most applications are only concerned with the model data. In both cases, these directories should contain a series of subdirectories with two-character names corresponding to the middle two characters of PDB IDs, each containing associated PDB files with specific names such as:

$PDB_MIRROR_PDB/hb/pdb1hbb.ent.gz
$PDB_MIRROR_STRUCTURE_FACTORS/hb/r1hbbsf.ent.gz

If the mirror is out of date and a specific entry cannot be found, the function will revert to downloading from the RCSB PDB site.

Installation

I'm upgrading to a new version of Phenix. How do I uninstall the previous version?

You can simply delete the entire folder, for example /usr/local/phenix-VERSION or (on Mac) /Applications/PHENIX-VERSION. On Mac, you can drag the folder into the Trash. Each installation is entirely self-contained and all project data is stored elsewhere (for instance in your home directory).

Can multiple installations of Phenix co-exist on the same computer?

Yes, but note that the command names will overlap, so you will either need to source the appropriate phenix_env script as needed, or launch programs with the version suffix:

phenix.refine-1.8.4-1496
phenix.refine-1.9-1962

Graphical interface

When I start up the Phenix GUI for the first time, I get an error message "ImportError: libjpeg.so.62: cannot open shared object file: No such file or directory". How do I fix this?

This error usually occurs on recent versions of Ubuntu, which do not install an essential library by default. You can fix it by running this command:

sudo apt-get install libjpeg62

How do I find a specific parameter in the GUI?

Open the Settings menu, and select Search all parameters. This can be used to search for any keyword in the parameter name, label, or help text.

Working with reflection files

What reflection file formats does Phenix support?

We recommend using MTZ files where possible but several other formats are also read (including XDS and Scalepack). Programs in Phenix rarely output any format other than MTZ, however. See the overview in the page for file formats for more details.

I have requested that the test set fraction be 0.05 (i.e. 5%), but Phenix only flags 1% of reflections. How do I make it use the number I specify?

By default there is a limit of 2000 on the number of reflections that will be picked for the test set; you don't really need more than this to ensure that R-free is statistically significant and that the maximum likelihood calculations work properly. However, if you want to use a full 5%, you can tell Phenix not to put any upper absolute limit. In the reflection file editor in the GUI, click "More options" in the R-free flags section of the editor, and delete the field labeled "Maximum number of reflections in test set".

Are MTZ files (or R-free flags) output by CCP4 data processing software suitable for use in Phenix?

Yes, absolutely. By default Phenix uses a different convention for labeling the test set when generating new R-free flags, but it will automatically detect the CCP4 convention and adjust as needed.

Phenix generated R-free flags with the test set denoted by 1, but REFMAC expects 0. How can I generate a more compatible MTZ file?

The reflection file editor has an option to export R-free flags to the CCP4 convention. Note however that newer versions of REFMAC will adjust to the Phenix convention automatically, so we recommend updating your CCP4 installation if you encounter this problem.

Should I use the MTZ file output by phenix.refine as input for the next round of refinement?

(Excerpted from the refinement FAQ, since this question is so common.)

The only time this is necessary is when you refined against a dataset that did not include R-free flags, and let phenix.refine generate a new test set. In this case, you should use the file ending in "_data.mtz" for all future rounds of refinement. You do not need to update the input file in each round, as the actual raw data (and R-free flags) are not modified.

Maps

I generated map coefficients in MTZ format using [some Phenix tool]. Why doesn't anything appear when I open it in Coot?

This almost always results from use of the "Auto-open map coefficients" menu item in Coot, which only recognizes several specific MTZ column labels output by programs such as phenix.refine, phenix.maps, Phaser, or REFMAC. (Several other programs in Phenix will now also output the FWT,PHWT columns that are interpreted as a 2mFo-DFc map, although depending on the program these may actually be a density-modified mFo map.) For any other map coefficient labels, for example an anomalous map, you will need to use the next menu item which requires explicit selection of the amplitude and phase labels. You can however modify the Coot preferences to recognize additional map types by editing the file ~/coot.py and adding one or more lines line the following:

set_auto_read_column_labels("FEM", "PHIFEM", 0)

which would make it possible to auto-open the feature-enhanced maps <../reference/fem.html>_. (Thanks to Georg Mlynek for this suggestion.)

Note that in most cases the Phenix GUI will automatically load the appropriate map coefficients for you upon clicking the Coot button in the results tab. Also note that on Macintosh, there are bugs in the X11/XQuartz software which often result in the window to select column labels appearing behind the main Coot window, which is easy to miss.

I have an MTZ file containing map coefficients. How can I obtain a map appropriate for display in PyMOL (or Chimera, etc.)?

In the Phenix GUI, this is almost always automatic: any time a button labeled "Open in PyMOL" is displayed, this will perform the necessary conversion to CCP4 format. In the phenix.refine GUI, there is also a separate button to simply generate the CCP4 maps without opening PyMOL. The map grid spacing, which controls the fineness of the displayed mesh, can be set in the "Graphics" section of the Preferences. By default this is a relatively conservative d_min/3 for reasons of speed and disk space. In most cases the map region will cover the atoms in the corresponding PDB file plus a generous buffer.

You may also create CCP4 (or X-PLOR or DSN6) maps manually. In the GUI, the program labeled "FFT map coefficients" in the "Maps" category performs this action; only an MTZ file is required as input, but a PDB file is recommended if you want to ensure that all atoms are covered. Every valid set of map coefficients in the MTZ file will be converted, with the exception of F(model) arrays. On the command line, this program exists as phenix.fft or phenix.mtz2map:

phenix.fft refine_001.mtz refine_001.mtz grid_resolution_factor=0.25

What is the best way to generate an omit map in Phenix?

There are several options (with some overlap in functionality):

• To remove model bias, the fastest option is to generate a composite omit map for the entire structure. By default this uses a very fast procedure for iteratively de-biasing the phases, but other (slower) options are available too.
• For omitting just a subset of atoms, an older and very popular option is to run phenix.refine with simulated annealing, here specifying an atom selection to be omitted. This also includes an option to harmonically restrain atoms to prevent other parts of the model from drifting into the omitted region.
• The composite omit map tool also has an option to use phenix.refine to de-bias the structure, in addition to the fast method described above. The refinement is much slower, but may be parallelized across multiple CPU cores or a cluster.
• Finally, the most thorough procedure for bias removal is the iterative build omit map option in AutoBuild. This rebuilds the entire model, in addition to running refinement.
• If you just want a simple omit map for a subset of atoms and do not care about model bias, phenix.maps has an option to omit a subset of atoms. This is suitable for quick calculations, but not for generating figures for publication.

Do Phenix maps contain R-free-flagged reflections?

In most cases, yes (except as noted below). For programs such as phenix.refine or phenix.maps, maps may optionall have the flagged reflections excluded, but this is not the default behavior. In the Phenix GUI, the "Refinement" section of the Preferences allows you to set this option this globally. Usually this will lead to flagged reflections being substituted by F(model) to prevent Fourier artifacts due to incomplete data; this may slightly increase model bias but it will not lead to biasing R-free if the resulting map coefficients are used for real-space refinement. The program phenix.remove_free_from_map can remove the flagged reflections from an existing MTZ file, and phenix.fft/pheniz.mtz2map has an option to perform this when converting to CCP4 or similar format.

Although the previous paragraph applies to output maps, the real-space refinement in phenix.refine always removes flagged reflections from the maps used internally. This is essential to prevent biasing R-free by refining against the real-space map incorporating the flagged reflections.

How do I apply B-factor sharpening or blurring to a map?

Both phenix.refine and phenix.maps allow you to customize the generation of map coefficients and/or CCP4 maps. In either case, there will be a button to enter map options, with several pre-defined map types already entered (at least 2mFo-DFc, mFo-DFc, and anomalous). Each map has a setting to activate map sharpening, and an input field to enter the B-factor to apply. Negative values sharpen the map, positive values blur it.

Note that Coot also allows you to perform these operations interactively (Calculate menu --> Map sharpening...), which may be more helpful than guessing a number.

How do I obtain an anomalous difference map in Phenix?

If you have anomalous data (separate F+ and F- or I+ and I-) and use this as input, both phenix.maps and phenix.refine will automatically output the anomalous difference map in the map coefficients MTZ file. Note that depending on the application, you may need to manually open the MTZ file in Coot to get the anomalous map to appear.

How can I calculate a log-likelihood gradient SAD map using Phaser?

Several programs in Phenix interface with Phaser to calcluate an LLG map given a model with existing anomalous (and real) scatterers. This map is more sensitive towards weaker scatterers (such as phosphorous, sulfur, and chloride, or partially-occupied heavy atoms) because it subtracts the contribution of existing atoms. (In this sense it is similar to any other difference map, where the sigma level of the map around remaining unmodeled atoms becomes increasingly higher as more atoms are added to the model). The first two options are recommended; in both cases you will need to request an additional set of map coefficients where map_type=LLG. (This is available as a preset option in the GUI.)

• In phenix.maps, if you specify the wavelength parameter, the entire structure will have scattering factors adjusted to model the anomalous scattering expected at the given wavelength. This is relatively fast, but the disadvantage is that the theoretical f'' may be either too high or too low, leading to difference map peaks around existing atoms.
• In phenix.refine, if you enable anomalous group refinement and specify the groups to be refined (or manually enter fixed f'' values), the LLG map will reflect the refined values rather than theoretical calculations. You can also set group_anomalous.find_automatically=True to identify appropriate groups automatically (based on atomic number).
• If you use the Phaser-EP GUI, there is also an option to output the LLG map coefficients at the end. However, if you turn on anomalous substructure completion, the final LLG map will usually be relatively flat. (Alternately, if you do not use substructure completion, you will be limited to whatever anomalously scattering atoms are present in the input sites file.) Therefore, unless you actually want to perform experimental phasing, when starting from a partial (or mostly complete) model the above methods are more useful.

Finally, for automation purposes, there is a utility for calculating a single set of map coefficients:

mmtbx.compute_map_coefficients model.pdb data.mtz wavelength=0.9792 map_type=llg

This should produce identical output to phenix.maps, but has been streamlined to require minimal input.

I have maps for the same protein(s) in different crystal forms, which I would like to compare. How can I superpose the maps along with the models?

There are two options for this, with slightly different limitations and uses:

• phenix.superpose_maps: given two models and two MTZ files containing map coefficients, superposes the models (using user-provided selections if desired), moves them into a P1 box (where all sites > [0,0,0]), then transforms the selected maps to follow. Only works on one pair of map coefficients at a time, but allows any type of complex array to be used.
• structure comparison: given M models containing N copies of chain X, and M corresponding MTZ files, extracts all copies of chain X, superposes in a P1 box, transforms the maps to follow. Works with any arbitrary number of models or chains, but limited to the most common map types (2mFo-DFc, mFo-DFc, anomalous).

Models

The validation GUI (or the PDB deposition annotator) is complaining about zero-occupancy atoms. Why does this happen, and how can I fix it?

The most common reason for atoms having zero occupancy is heteroatom groups (in this context, any atom labeled as HETATM in the PDB file) being reset to zero occupancy by Phaser. Although this can be avoided by modifying the Phaser settings, it is easy to fix using several different programs:

• In PDBTools, there is an option to reset all selected occupancies to a specified value (usually 1.0). If you do not want any non-unity occupancies it is safe to do this for the entire model, otherwise you will need to enter an atom selection to modify.
• For more manual control, the PDB file editor in the Phenix GUI allows you to change the properties of individual atoms (or residues, chains, etc.).
• In phenix.refine, there are options to modify the starting model in the middle panel of the configuration interface; these are essentially identical to the options in PDBTools. (On the command line, the parameter scope "modify_start_model" contains these options.)

Please note that any time you modify the atomic properties (or add or remove atoms), it is essential to refine the model again, as the previously calclulated R-factors and maps will be invalid.

What is the difference between ATOM and HETATM records in a PDB file?

Within Phenix, few distinctions are made between the different atom labels. ATOM generally refers to canonical amino acid residues or nucleic acid bases; HETATM ("heteroatom") covers everything else, including non-canonical amino acids such as selenomethionine (MSE), covalently modified residues, ligands, and solvent. However, this distinction is not enforced (and is frequently ignored) within Phenix, and the choice of label will usually have no effect on program behavior. (The lone exception is Phaser, which by default sets the occupancies of HETATMs to zero unless they are part of a protein polymer.) The PDB annotators will change ATOM to HETATM as needed upon deposition.

Rosetta

(These questions pertain to the Rosetta software for protein structure prediction and design, which is developed and distributed separately from Phenix but is used internally by tools such as MR-Rosetta and phenix.rosetta_refine.)

When I try to build the hybrid Python-enabled Rosetta for running refinement, I get the error below. What does it mean and how can I fix it?

[root@localhost ~]# rosetta.build_phenix_interface nproc=8
scons: Reading SConscript files ...
Running versioning script ... Done. (0.0 seconds)
Could not find platform independent libraries <prefix>
Could not find platform dependent libraries <exec_prefix>
Consider setting $PYTHONHOME to <prefix>[:<exec_prefix>]
ImportError: No module named site

This error happens when the Rosetta build system tries to run Python internally to generate additional files used in compilation. Although we include a complete Python environment with Phenix, Rosetta will probably use the system's Python interpreter (or whatever is defined as the 'python' command). The error message you see means that this interpreter is completely broken, i.e. something is wrong with your system. You can confirm this by running a command such as:

python -c "import site"

You should consult your local system administrator and/or OS reference manual for assistance fixing the problem.

Other

How can I obtain phases and Hendrickson-Lattman coefficients for my model and data?

There is a program specifically for this purpose; you can find in in the Phenix GUI under "Reflection tools".