Rosetta 3.3
Documentation for ca_to_allatom application


Last edited Oct 31 2010 by Frank DiMaio. Code by Frank DiMaio ( P.I.: David Baker.

Code and Demo

The Mover for this application is in src/protocols/RBSegmentMoves/ Sub-movers that may be useful are also found in this folder. This code is also used as part of a tutorial on using Rosetta to refine structures into density. The complete demo is in demo/electron_density/cryoEM_tutoral. The portion of the tutorial using this application is in 'scenario6_model_from_ca_trace'.



This is an application intended to aid in building a full-atom mdoel from a low-resolution C-alpha-only trace (such as hand-traced models from cryo-electron microscopy). The input to the protocol includes the initial C-alpha trace and a rigid-body segmentation file, which identifies secondary-structure elements in the initial trace.


The protocol is divded into 4 phases:

The first two phases may be done in either all-atom or centroid mode. Loop modelling is always done in centroid mode, while all-torsion minimization is always fullatom.


This code assumes the starting C-alpha model contains some errors. If you want a reasonable backbone model with exactly the C-alpha coordinates specified a geometric tool like MaxSprout may be more appropriate.

Input Files

The input PDB -- even though it only needs coordinates for C-alpha atoms -- must have lines for all backbone atoms. These may have coordinates of all 0's, with an occupancy set to -1, or a tool like MaxSprout can be used to find some geometrically reasonable conformation. These coordinates are not used at all by rosetta, but the PDB-reading machinery requires their presence.

The RB segment file (specified with -RBSegmentRelax:rb_file) is as follows:

SEGMENT   1   4   8  E
SEGMENT   2  13  16  E
SEGMENT   3  26  34  E

The first column is the word 'SEGMENT', the second is a unique identifier for each segment, the third and fourth a residue range, and the fifth is 'E', 'H', or 'X' depending on secondary structure type (X means compound, for example, if modeling a strand-turn-strand motif as a single element).

A vall file (specified with -loops:vall) is used to generate backbone fragments dynamically. A version of it is included in the Rosetta database.


The options listed below include those specific to the ca_to_allatom application and a selection of common flags that may be useful. If a post-run looprelax is enabled, then the app takes all looprelax and relax specific options. If modelling into density, the app takes all density-fitting options.

Common options:

These flags control whether fullatom or centroid models are used during the first two phases of the protocol (defaults to fullatom). The scorefunction given must agree with this.

The rigid-body segmentation file. See notes above for more info.

The weight and width of the constraints during the first two phases of the protocol. Wider and weaker constraints will produce structures further from the initial CA trace.

The number of moves in phase 1 and phase 2 of the protocol.

The scorefunction used during the first 2 phases of the protocol. If the input pose is fullatom, make sure this is a fullatom function (like score13_env_hb); if centroid make sure it is a centroid function (like score5).

This tell the protocol to skip one of the three move types. Note that if fragment moves are skipped, only 1 fragment insertion is done for each secondary structure element (to build the initial full-atom pose from a C-alpha trace).

The movement parameters, specified as angle-distance pairs. The helical and strand moves have two pairs of these parameters, corresponding to on-axis and off axis movements, while the default case has only one pair. As an example, '-strand_movement_params 40.0 1.0 1.0 0.2' will perturb strands 40 degrees/1A along the strand axis, and 1 degree/0.2A outside the strand axis. These parameters default to 0, so must be specified by the user.

Should we skip looprelax?

Rarely-used options:

If specified, all ligands in the inital trace will not move at all from the starting position

When dynamically choosing fragments in phase 1, this parameter determines how much randomization goes into the search. In general, higher values for this will result in structures further from the starting model.


Post Processing

Post-processing is highly dependent on the accuracy of the initial C-alpha trace. If the lowest-energy models show good convergence (either by visual inspection or clustering) then minimal postprcessing is needed. Alternately, these models may serve as starting points for additional modelling with loopmodeling.

 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines