Sequence assignment and linkage of neighboring segments with assign_sequence

Author(s)

Purpose

You can now carry out an improved sequence assignment of a model that you have already built with phenix.assign_sequence. Further, once the sequence has been assigned, this method will use the sequence and proximity to identify chains that should be connected, and it will connect those that have the appropriate relationships using the new loop libraries available in phenix.fit_loops. The result is that you may be able to obtain a more complete model with more chains assigned to sequence than previously.

assign_sequence is a command line tool for reanalyzing resolve sequence assignment for a model and a map including the non-crystallographic symmetry, exclusion of sequence by previously-assigned regions, and requirement for plausible distances and geometries between ends of fragments with assigned sequences. Additionally assign_sequence will use the fit_loops loop library to connect segments that are separated by a short loop.

Note: assign_sequence is designed to be used after resolve model-building in which residues that are not assigned to sequence are given residue numbers higher than any residue in the input sequence file. If you input a model not built by resolve or in phenix, or if you would like to completely redo the sequence assignment for your model, be sure to set "allow_fixed_segments=False".

NOTE: assign_sequence is normally called from phenix.phase_and_build but you can run it interactively if you want.

Usage

How assign_sequence works:

The starting point for assign_sequence is a set of segments of structure read in from the input model. assign_sequence then uses resolve to calculate the compatibility of each possible side chain with each residue in each segment. Then assign_sequence tests out possible combinations of alignments of all the segments in the input model and chooses the set of alignments that is most compatible with the density map, the number of NCS copies, and with the geometries and distances between ends of the segments.

Sequence probabilities:

assign_sequence uses the side-chain to map compatibility matrix calculated by resolve to assess the relative probabilities of each possible side chain at each position in the input model. Segments that are positively assigned to a sequence by resolve are (by default) maintained and used as anchors for further sequence assignment. All other segments have a relative probability associated with each possible alignment of the segment to the input sequence. The score for each alignment is the logarithm of this probability (essentially a log-likelihood LL score).

Connection scores:

Any pair of segments with some assignment of sequence to each segment has an additional score corresponding to the plausibility of a connection of the expected length existing between the segments. If the distance between ends is greater than can be bridged by the number of residues separating them, then the connection is not possible. If the connection is possible, it is scored based on the best density fit (CC) of a loop from the fit_loops loop library. This additional score is normally 10*CC.

Generating sequence alignments and connectivities

assign_sequence starts with the segments with the most convincing assignments of sequence. Often these are those with sequence positively assigned by resolve; otherwise they are those with the highest-probability assignments. This yields a starting arrangement (sequence assignment for a set of segments). Then each possible sequence assignment of each unassigned segment is tested for compatibility with the existing arrangement and the one that is most compatible (based on the connections that would result, duplication of sequence, and sequence-map matching) is added to the arrangement. Optionally many arrangements can be built up in parallel, but often a very good one can be found simply by taking the top one at each step. This process is repeated until no additional segments can be added to the arrangement to yield an increase in log-likelihood score of (by default) 2 or greater.

NCS copies:

assign_sequence builds up a set of possible sequence assignments and connectivities that depends on the expected number of copies in the asymmetric unit of the crystal. If there is only one copy of the molecule in the crystal, then no residues in the sequence can be used more than once in sequence assignment. If there are N copies, then a residue can be used up to N times. If there are multiple copies, then each molecule must be self-consistent, with plausible distances and geometries relating each segment to the next.

Connecting segments:

Once a final arrangement is found, including NCS if applicable, all segments that are separated by short loops (typically 0-3 residues) are connected using loops from the fit_loops loop library. This yields longer segments of structure with sequences fully assigned. The resulting model then has side chains added to match the newly-assigned sequence and is written out.

Output files from assign_sequence

assign_sequence.pdb: A PDB file with your input model assigned to sequence (to the extent possible). Residues not assigned to sequence will be given a chain ID higher than those assigned, and they will be given residue numbers higher than any residue number in the sequence file.

Examples

Standard run of assign_sequence:

Running assign_sequence is easy. From the command-line you can type:

phenix.assign_sequence map_coeffs.mtz coords.pdb sequence.dat

If you want (or need) to specify the column names from your mtz file, you will need to tell assign_sequence what FP and PHIB (and optionally FOM) are, in this format:

phenix.assign_sequence map_coeffs.mtz coords.pdb \
labin="FP=2FOFCWT PHIB=PH2FOFCWT" sequence.dat

Possible Problems

Specific limitations and problems:

Literature

Additional information

List of all available keywords