Tool: Modeller Comparative

The Modeller Comparative modeling tool is an interface to the Modeller program, either installed locally or run on a web service hosted by the UCSF RBVI. The command-line implementation is modeller comparative. See also: Model Loops, Rotamers, Build Structure, AlphaFold, Task Manager, esmfold

Comparative modeling (sometimes called “homology modeling”) generates theoretical models of a protein using one or more known related structures and a sequence alignment of the known and unknown structures. The protein chain to be modeled is the target, and a known atomic structure used for modeling is a template.

Modeller is developed by the Šali Lab, and its users should cite:

Comparative protein modelling by satisfaction of spatial restraints. Šali A, Blundell TL. J Mol Biol. 1993 Dec 5;234(3):779-815.

Modeller comparative modeling requires the following for each target chain:

1. At least one template structure.

   Modeling a multimer requires a multimeric template structure with the same stoichiometry and expected spatial relationship. Modeller uses this stoichiometry and spatial relationship rather than trying to calculate it.

   
   
2. A sequence alignment that includes the sequence of the target; other sequences may also be included. The sequence alignment and template structure(s) should be open in ChimeraX and the template(s) associated with sequence(s) in the alignment.

   The target sequence can be fetched from UniProt or opened from a file. That single sequence suffices as the “alignment” if the template is similar enough in sequence to associate with it (see an example). Otherwise, it can be used as a query to search the PDB for possible template structures using Blast Protein (which can be started from the Sequence Viewer context menu). Alternatively, a multiple sequence alignment of the target and template(s) can simply be opened from a file, if available.

   Modeling a heteromultimer requires a separate sequence alignment for each unique chain. For example, modeling an α2β2 tetramer requires a template structure that is also a α2β2 tetramer with its two α subunits associated with one alignment containing the target α sequence, and its two β subunits associated with another alignment containing the target β sequence.

3. A Modeller license key. Academic users can register free of charge to receive a license key. Commercial entities and government research labs, please see Modeller licensing.

[back to top: Modeller Comparative]

Startup and Input

The Modeller Comparative tool can be opened from the Sequence or Structure Prediction section of the Tools menu and can be manipulated like other panels (more...). It can also be started from the Sequence Viewer context menu.

One sequence alignment per target should be chosen (highlighted) in the top section of the dialog. Immediately below, for each chosen alignment, the name of the target sequence should be selected from the pull-down menu.

Basic settings:

• Make multichain model from multichain template (initial default on) – whether to retain template homomultimeric states; for example, if the template structure is a homotrimer, whether to model the target as a homotrimer instead of as a monomer with three templates
• Modeller license key – see above
• Number of models (initial default 5, maximum 1000) – number of models to create
• Computation location
  • web service (initial default) – run the Modeller web service hosted by the UCSF RBVI; the user must provide:
    • Modeller license key – see above
  • local machine – run a local installation of Modeller; the user must install Modeller version 10 or newer, then provide:
    • Executable location – the pathname of the Modeller executable

Advanced settings:

• Build models with hydrogens (initial default off) – whether to include hydrogen atoms in the output models; warning: slow
• Include non-water HETATM residues from template (initial default off) – whether to include HETATM residues other than water (ligands, ions, detergent, etc.) from templates in the output models. Any such residues not desired in the output should be deleted from the template(s) beforehand if the option is turned on.
• Include water molecules from template (initial default off) – whether to include water residues from templates in the output models. Any water residues not desired in the output should be deleted from the template(s) beforehand if the option is turned on.
• Temporary folder location (optional) – use the specified location for temporary files; otherwise, a location will be generated automatically
• Use fast/approximate mode (produces only one model) (initial default off) – use fast/approximate mode (~3 times faster) to get a rough idea of model appearance or to confirm that the alignment is reasonable. This mode does not randomize the starting structure (generates only a single model) and performs very little optimization of the target function.

Modeller Comparative settings including the license key are automatically saved in the preferences. Other than the license key, the remembered settings apply only to the GUI; the command defaults are not changed.

Clicking Cancel simply dismisses the dialog, whereas OK sends information to the web service and initiates the calculation.

[back to top: Modeller Comparative]

Modeller Results

When results are returned, the new models are opened, listed in the Model Panel along with any other models present in ChimeraX, and automatically superimposed on the lowest-model-ID template structure (for comparative modeling) or the original structure (for loop modeling and refinement) with matchmaker. Scores for the models are shown in a Modeller Results panel:

• Model – model ID number
• GA341 – model score derived from statistical potentials (see Melo et al., Protein Sci 11:430 (2002)), automatically assigned as a model attribute named modeller_GA341. Higher values signify better models, with those > 0.7 generally indicating a reliable model, defined as ≥ 95% probability of correct fold.
• zDOPE – normalized Discrete Optimized Protein Energy (DOPE), an atomic distance-dependent statistical score (see Shen and Sali, Protein Sci 15:2507 (2006)), automatically assigned as a model attribute named modeller_zDOPE. Lower values signify better models, with those < 0 generally indicating a reliable model.

Clicking a row in the panel shows the corresponding model and hides the others. Clicking elsewhere in the panel shows all of the models at once. Clicking a column header sorts on the values in that column.

The panel's context menu includes an option to Fetch Additional Scores using the SaliLab Model Evaluation Server:

• GA341 and zDOPE (described above), if missing for any models
• predicted NO35 – TSVMod-predicted native overlap (3.5 Å), the fraction of Cα atoms in the model within 3.5 Å of the corresponding atoms in the native structure after rigid-body superposition (see Eramian et al., Protein Sci 17:1881 (2008)); automatically assigned as a model attribute named modeller_predicted_NO35.
• predicted RMSD – TSVMod-predicted Cα root-mean-square deviation (RMSD) of the model from the native structure (see Eramian et al., Protein Sci 17:1881 (2008)); automatically assigned as a model attribute named modeller_predicted_RMSD.

The same Modeller license key as needed for comparative or loop modeling is also required by the evaluation server, but it does not need to be specified again. Fetching additional scores is also implemented as the command modeller scores.

After additional scores have been fetched, the context menu option changes to Refresh Scores to allow re-evaluation after models have been modified (for example, to delete untemplated regions built as extended tails). However, since re-evaluation does not use all of the same information as does scoring during the original modeling process, it may worsen pre-existing GA341 and zDOPE scores and should only be used after making modifications that are expected to improve the scores.