An automated model building pipeline for X-ray crystallography

Getting Started

ModelCraft can be installed using pip, e.g.

python3 -m pip install --user modelcraft

Refer to the pip documentation if pip is not installed. ModelCraft also requires an installation of CCP4. The CCP4 environment needs to be set up so that programs such as Buccaneer and Refmac can be called from the command line. The simplest execution requires only a reflection data file in MTZ format (with observations, a free-R flag and starting phases) and a description of the asymmetric unit contents.

modelcraft --data data.mtz --contents contents.json

Alternatively, a model can be provided (in PDB or mmCIF format), which will be refined and used as a starting point instead of starting from phases in the data file.

modelcraft --data data.mtz --contents contents.json --model model.cif

Required Arguments

--data FILE
Input reflection data in MTZ format. This must contain observations and a free-R flag. If a starting model has not been provided then it must also contain phases as a phase and figure-of-merit or Hendrickson-Lattman coefficients. An attempt will be made to automatically identify column labels unless they are specified using the --observations, --phases and --freerflag arguments.

--contents FILE
A description of the assymetric unit contents, either as a sequence file (in FASTA or PIR format) or a more detailed contents file in JSON format. See the ASU Contents Description section of the documentation for more details.

Optional Arguments

Run a basic pipeline using only Buccaneer and Refmac. Parrot density modification is still used on the first cycle and input models are still refined using Sheetbend and Refmac.

--convergence-cycles N default: 4
Number of cycles without improvement needed to stop the pipeline automatically. --convergence-tolerance is used to determine whether a cycle gave an improvement.

--convergence-tolerance default: 0.1
R-free difference needed to mark a cycle as an improvement. If the model at the end of a cycle has an R-free that is this much better than the previous best R-free then the cycle is marked as an improvement. --convergence-cycles states how many cycles without improvement are needed to stop automatically.

--cycles N default: 25
Maximum number of pipeline cycles.

--freerflag COL e.g. FreeR_flag
Column label for the free-R flag.

Show an automatically generated help message.

Keep the files from intermediate jobs instead of deleting them. This can lead to large directory sizes.

Keep log files if not using the full --keep-jobs argument.

--model FILE
Starting model. If input phases are not specified this will be refined using Sheetbend followed by Refmac.

Run the maximum number of cycles even if the model is not improving.

--observations COLS e.g. FP,SIGFP
Column labels for the observations as either amplitudes or intensities (both can be mean or anomalous).

--phases COLS e.g. PHIB,FOM or HLA,HLB,HLC,HLD
Column labels for input phases as either a phase and figure of merit or Hendrickson-Lattman coefficients.

Turn on twinned refinement. Only do this if you are sure your data are twinned.

Pass input phases to Refmac for MLHL refinement.

Developer Arguments

--buccaneer FILE
Path to an alternative buccaneer executable.

--parrot FILE
Path to an alternative parrot executable.

--sheetbend FILE
Path to an alternative sheetbend executable.

ASU Contents Description

A description of the expected contents of the asymmetric unit must be provided as a FASTA sequence file or a JSON file using the --contents argument. A sequence file is simpler, but the JSON format has the following advantages:

In order to create a JSON file it may be helpful to start from the contents for an existing PDB entry. The modelcraft-contents script creates a contents JSON file for a released PDB entry.

An example JSON file is shown below:

    "copies": 2,
    "proteins": [
            "stoichiometry": 1,
            "modifications": ["M->MSE"]
            "start": 1,
            "sequence": "GGG"
    "rnas": [
            "sequence": "GGUAACUGUUACAGUUACC",
            "stoichiometry": 2,
            "modifications": ["5->GTP", "23->CCC"]
            "start": 5,
    "dnas": [],
    "carbs": [
        { "codes": { "NAG": 2 }, "stoichiometry": 1 },
        { "codes": { "MAN": 1, "NAG": 2 }, "stoichiometry": 1 }
    "ligands": [
        { "code": "HEM", "stoichiometry": 1 }
    "buffers": ["GOL", "NA", "CL"]

The file has a list of proteins, rnas, dnas, carbs, ligands, and buffers that are in the crystal. The only mandatory items are that each protein, RNA or DNA chain must have a sequence, each carbohydrate must have a dictionary of codes to specify the number of each sugar, and each ligand must have a single code.

Each component (other than buffers) has a stoichiometry parameter to specify the stoichiometry. In the example above there are 2 RNA chains for each protein chain. If the stoichiometry is not specified it is assumed to be 1. There is also a copies parameter for the whole file to specify how many copies of the contents are in the asymmetric unit. If this value is not known the most likely number will be estimated. The modelcraft-copies script can be used to view the solvent fraction and probability for each number of copies given a contents file and an MTZ file. It is assumed that the number of ordered buffer molecules is unknown so they are not included in the solvent calculation.

Finally, protein, RNA and DNA chains may have a list of modifications, e.g. M->MSE to specify that all methionine residues are actually selenomethionine or 5->GTP to specify that the residue 5 is guanosine triphosphate. The start property changes the residue numbers in modifications, e.g. residue 5 in the RNA chain is actually the first residue.

Note: ModelCraft does not yet build carbohydrates, ligands, or modified residues (other than selenomethionine derivatives). However, this is planned for the future and inclusion of these components in the contents allows for more accurate calculation of the solvent fraction during density modification.


P Bond. Next generation software for placing atoms into electron density maps. PhD thesis, University of York (2021) URL
K Cowtan. Acta Cryst. D, 62, 1002 (2006) DOI
P Emsley, B Lohkamp, WG Scott, K Cowtan. Acta Cryst. D, 66, 486 (2010) DOI
K Cowtan. IUCrJ, 1, 387 (2014) DOI
K Cowtan. Acta Cryst. D, 66, 470 (2010) DOI
O Kovalevskiy, RA Nicholls, F Long, A Carlon, GN Murshudov. Acta Cryst. D, 74, 215 (2018) DOI
K Cowtan, S Metcalfe, P Bond. Acta Cryst. D, 76, 1192 (2020) DOI