Ensemble creation with Ensembler

Contents

Purpose

Ensembling is a superposition of multiple homologous structures, which often gives superior performance when used as search model for molecular replacement.

Usage

There is a standalone ensembler_new program available from the command line. The unified model preparation program sculpt_ensemble is available from the command line and also from the PHENIX GUI.

Input files

Output files

The superposed chains can be written out either as a quasi multiple model PDB file that is readable by phaser directly (output style merged, output file name root_merged.pdb) or as a series of files containing each chain separately (output style separate, file name root_pdb_chain.pdb, where pdb is the name of the PDB file the chain was read from, and chain is the chain identifier). The file name root can be changed via the root parameter of the output (default: ensemble).

Description

The workflow consists of several stages that can be independently configured. These are listed in order of execution. For a summary of all parameters with the corresponding defaults, see the Additional information section.

Model input

The syntax allows the definition of structural assemblies in a flexible manner, and is also backwards-compatible with the previous version, i.e. existing command lines do not have to be changed.

Single-chain structures

These only require the input of files that contain the structures to be superposed. In case multiple chains are present in a file, these are superposed independently, but it is also possible to restrict participating chains to a selection (using the selection keyword).

For example, to superpose all chains in a.pdb with all chains in b.pdb, the following PHIL is needed (these can also be specified as command-line arguments):

input.model.file_name=a.pdb
input.model.file_name = b.pdb

To superpose only chains A and B from a.pdb with chains A and C from b.pdb, the selection keyword can be used (please note that this can now longer be achieved from the command line):

input {
  model {
    file_name = a.pdb
    selection = chain A or chain B
    }

  model {
    file_name = b.pdb
    selection = chain A or chain C
    }
  }

Multimeric assemblies

In this case the individual assemblies need to be specified using either the assembly keyword or the read_single_assembly_from_file keyword if the respective structure file contains a single assembly only.

In case the input PDB file contains multiple MODEL sections, these are automatically treated as separate assemblies and no further action is required. However, it is still possible to omit chains from the assembly by listing only the chainIDs on the assembly keyword. On the other hand, the read_single_assembly_from_file is not applicable to such a PDB file, and will result in an error message.

For example, the structure 4QTR contains two copies of a protein-DNA complex: protein chains A and B interact with DNA chains E and F, while protein chains C and D interact with DNA chains G and H.

To superpose these two assemblies, the following input is required:

input {
  model {
    file_name = 4qtr.pdb
    assembly = A B E F
    assembly = C D G H
    }
  }

To superpose a subset of the assemblies, it is sufficient to specify the subset to be superposed. Additional chains are automatically discarded. For example, to superpose a partial assembly from the previous example, one can use the following command file:

input {
  model {
    file_name = 4qtr.pdb
    assembly = A E F
    assembly = C G H
    }
  }

If the assemblies to be superposed are present in separate files, one can use a more convenient syntax:

input {
  model {
    file_name = 4qtr_1.pdb
    read_single_assembly_from_file = True
    }
  model {
    file_name = 4qtr_2.pdb
    read_single_assembly_from_file = True
    }
  }

Input summary

The program list the chains that were read from input files, and also displays the identified macromolecule type. It also does a consistency check to ensure that all assemblies have the same number of chains and these are of identical types.

Residue mapping

Establishes the equivalence of residues among the input chains. There are several options available:

  • ssm - Do structure-based pairwise SSM alignment, using the first chain as reference (default).
  • ssm_multiple_alignment - Do structure-based multiple SSM alignment.
  • muscle - Generate a sequence alignment with MUSCLE and use that to align the residues.
  • multiple_alignment - A user supplied multiple alignment is used. An error message will be raised if this is not present or does not cover all present chains.
  • resid - Residue sequence number with insertion code will be used.

Residue mapping can be requested on an assembly-member basis, by specifying the requested options in the same order as the chains of the assembly. This is not compulsory, and if no options are specified, a default dependent on the macromolecule type will be used (for protein, the default is ssm, for nucleic acids it is muscle). It is also possible to request specific mapping algorithms for a subset of all chains of the assembly, but in this case these have to come first in the list.

For example, to request non-default mapping options for the 4QTR example, the following options are needed (input section repeated for convenience):

input {
  model {
    file_name = 4qtr.pdb
    assembly = A B E F
    assembly = C D G H
    }
  }

configuration {
  mapping = ssm muscle muscle resid
  }

This will use the ssm option for chains A and B, muscle for chains B and D, muscle for chains E and G and resid for chains F and H. Note that these coincide with the defaults for chains A and B (assembly position 1, protein chain) as well as for chains E and G (assembly position 3, DNA chains). To request non-default options only, one has to rearrange the assembly so that positions 2 and 4 in the current setup become positions 1 and 2:

input {
  model {
    file_name = 4qtr.pdb
    assembly = B F A E
    assembly = D H C G
    }
  }

configuration {
  mapping = muscle resid
  }

In this case, since no specific mapping is requested for assembly positions 3 and 4, the default options will be used.

Atom mapping

Maps selected atoms within equivalent residues to each other. The mapping is done by name hence the order of atoms in the residue does not matter. If atoms are missing from certain residues (or if certain residues contain extra atoms), a gap will be filled where necessary.

Atom selection is controlled by the atoms parameter of the configuration scope. If multiple atom names are requested, these need to be separated by a comma (,). If no selection is requested, a type dependent on the macromolecule will be used (CA for protein and P for nucleic acids). There are also type-dependent shorthands available (backbone and all).

For assembly structures, it is possible to request atom selection on an assembly-member basis, with the same limitation that was explained in the previous section. As an example, here is a selection for the 4QTR structure:

input {
  model {
    file_name = 4qtr.pdb
    assembly = A B E F
    assembly = C D G H
    }
  }

configuration {
  atoms = C,CA,N,O backbone all backbone
  }

This instructs the program to use atoms C, CA, N, O for assembly position 1 (chains A and C), backbone atoms for assembly position 2 (chains B and D; since these are protein chains, the shorthand backbone stands for atoms C, CA, N and O), all atoms for assembly position 3 (chains E and G ) and backbone atoms for assembly position 4 (chains F and H; since these are DNA chains, backbone selection includes all atoms from the phosphate and the deoxy-ribose).

Superposition

Equivalent positions are superposed iteratively to find a globally optimal solution. There are two superposition algorithms implemented, which primarily differ in how they handle gaps in equivalent positions.

  • gapless. This algorithm discards all positions where gaps are present, For further details, see Diamond (1992).
  • gapped. This algorithm can use all positions as long as there are two sites presents (i.e. not gaps), and may give better results for chains with distant sequence identity. For the exact algorithm, see Wang & Snoeyink (2008).

Both algorithms use Diamond's formulation to solve the pairwise rotational superposition problem (Diamond, 1988).

An exception is raised if there are less than 3 sites present for superposition.

Multiple superposition is an iterative process and consists of a series of pairwise superpositions. The convergence criterion is controlled by the convergence parameter (in the superposition scope), which is the r.m.s. difference change between two consecutive iterations.

Weighting

Automatic weighting can be used to improve superposition, either to amplify highly homologous regions or to decrease the effect of incorrect site-equivalence (typically arises because of a wrong alignment). Implemented weighting schemes are as follows:

  • unit - Unit weights (equivalent to no weighting).

  • robust_resistant - Robust-resistant weighting scheme (default). This tends to converge fast and give reliable results. The exact formula for weighting is as follows:

    w = 1 - ( delta2 / tolerance2 )2 if delta < tolerance
    w = 0 (otherwise)
    

    where delta is the deviation from the average. Tolerance is an empirical value, and its optimal value is close to (unweighted r.m.s.d.) 2 and can be controlled by the critical parameter of the robust_resistant scope.

Weighting is iterated with superposition until weights converge, which can be controlled by the convergence parameter of the weighting scope.

In case of highly dissimilar structures (or incorrect residue mapping), weight determination may temporarily need to be damped to avoid divergence. This is done automatically (in steps controlled by the incremental_damping_factor parameter of the weighting scope), until a preset value (controlled by the max_damping_factor parameter of the weighting scope) is reached, at which point an exception is raised.

Cluster analysis

Hierarchical cluster analysis is performed using the pairwise r.m.s. differences as a distance measure. The clustering parameter of the configuration score can be used to adjust cluster boundaries. Clustering information can be requested for multiple thresholds. Default: no clustering performed.

Chain trimming

This option trims residues from the final superposed model where the unweighted r.m.s.d. is above a certain threshold (threshold parameter of the trimming scope). Useful in removing flexible loops, etc. Default: no trimming.

Sorting

After superposition is complete, the chains can be sorted by sequence identity (identity), fraction of common sites wrt all aligned atom positions (overlap), weighted r.m.s.d. (wrmsd) or unweighted r.m.s.d. (unwrmsd). This is controlled by the sort parameter of the output scope. Default: input order (input).

Command line

phenix.ensembler \
    [ command-line switches ] \
    [ PHIL-format parameter files ] \
    [ PHIL command-line assignments ] \
    [ PDB-files ] \
    [ alignment files ]

Command-line switches

-h, --help            show this help message and exit
--show-defaults       print PHIL and exit
-i, --stdin           read PHIL from stdin as well
-v, --verbosity       set verbosity level (info,debug,verbose)
--version             show program's version number and exit
--text-logfile FILE   Verbatim copy of log stream
--html-logfile FILE   Verbatim copy of log stream in HTML

PHIL arguments

Everything not starting with a dash('-') is interpreted as a PHIL argument. This can be a PHIL-format file containing parameters, command-line assignment or a file whose type is automatically recognized (based on file extension; structure files and alignment files are recognized automatically).

Specific limitations and possible problems

Processing features

  • When an alignment-based residue mapping protocol is requested, very short residue segments (shorter than 3 consecutive residues) cannot be reliably aligned to the sequence, and these will be discarded from the superposition.

Warning and error messages

  • No input pdb files: no PDB files were given.
  • Cannot treat multiple MODELs as a single assembly: the read_single_assembly_from_file keyword was used in connection with a PDB file that contains multiple MODEL records.
  • No chains read from input PDB files: no superposable chain was found in input files. Superposable chains are protein, DNA and RNA.
  • No chain with allowed type found: no superposable chains found in structure file. Superposable chains are protein, DNA and RNA.
  • No chain '..' found: chain '..' (requested as part of an assembly) is not present.
  • Inconsistency between assembly 1 and assembly ..: different number of chains: assembly 1 and assembly .. do not have the same number of chains.
  • Inconsistency between assembly 1 and assembly ..: different chain types: an assembly position in assembly 1 and assembly .. do not have the same macromolecule type. The information on chains types in the assembly is given prior to this error message.
  • Mapping .. is not applicable to chain type ..: the mapping method .. is not applicable to a chain with macromolecule type .. (e.g. SSM superposition can only be used with protein chains).
  • No alignment specified: no alignment file was specified and a user-supplied alignment based mapping method was requested.
  • No matching alignment found: no alignment sequence matches from the ser-supplied alignment matches the chain sequence.
  • SSM error:...: an error occurred during SSM superposition.
  • No sites selected: the requested atom selection results in no sites selected for superposition.
  • Less than 2 chains for superposition: superposition cannot proceed because there are less than 2 chains found.
  • Insufficient overlap among structures: superposition cannot proceed
    because there are less than 3 sites available (this is checked after discarding gaps). If this error occurs, discarding the chain with the lowest overlap (printed just before the error is raised) may allow one to proceed.
  • Excessive weight shift, damping weight change: this warning is emitted if weights need to be damped, because all weights are zero.
  • Weight damping recovery exhausted: this error is raised if even after successive rounds of re-weighting, all weights are still zero. This either indicates incorrect alignment of the chains or too tight expected error.

References

[Diamond1988]A note on the rotational superposition problem R. Diamond Acta Cryst. A44, 211-216 (1988)
[Diamond1992]On the multiple simultaneous superposition of molecular structures by rigid body transformations R. Diamond Protein Science 1, 1279-1287 (1992)
[WangSnoeyink2008]Defining and computing optimum RMSD for gapped and weighted multiple-structure alignment X. Wang and J. Snoeyink EEE/ACM Transactions on Computational Biology and Bioinformatics 5, 525-533 (2008)

Additional information

List of all available keywords

  • inputInput files
    • modelInput model file
      • file_name = None Input file name
      • selection = None Selection string
      • assembly = None Assembly information
      • read_single_assembly_from_file = False Consider contents of the file as a single assembly
  • outputOutput file(s)
    • location = ""
    • gui_output_dir = None Sets base output directory for Phenix GUI - not used when run from the command line.
    • root = ensemble Output file root
    • style = *merged separate Output options for ensemble
    • sort = wrmsd *input cluster overlap unwrmsd Sort ensemble components
    • error_file_root = None Output file root for superposition error files
    • retain_associated_chains = False Keep (and transform) chains that are not assembly members
    • job_title = None Job title in PHENIX GUI, not used on command line
  • configurationGeneral parameters for ensemble generation
    • mapping = None Residue mapping methods. Valid choices: ssm, multiple_alignment, ssm_multiple_alignment, muscle, resid
    • alignment = None Alignment file (only necessary for 'alignments' modes)
    • atoms = None Atom to include in superposition
    • clustering = None Cutoff distance for cluster analysis
    • superpositionSuperposition setup
      • method = *gapless gapped Superposition algorithm
      • convergence = 1.0E-4 Convergence criterion for superposition
    • weightingParameters for weighting scheme
      • scheme = unit *robust_resistant Weighting scheme
      • convergence = 1.0E-3 Convergence criterion for weight iteration
      • incremental_damping_factor = 1.5 Damping factor in recovery cycle
      • max_damping_factor = 3.34 Quit recovery if cumulative damping factor is above
      • robust_resistantSetting for robust-resistance scheme
        • critical = 3 tolerance
    • trimmingTrimming parameters
      • trim = False Trim differing loops
      • threshold = 3.0 Max position RMSD for inclusion