StarMap: cryoEM Refinement with Rosetta
StarMap: a user-friendly workflow for Rosetta-driven molecular structure refinement.
Lugmayr W, Kotov V, Goessweiner-Mohr N, Wald J, DiMaio F, Marlovits TC.
Nat Protoc. 2023 Jan;18(1):239-264.
- Starmap is a ChimeraX plugin that provides an
easy-to-use (GUI) interface to Rosetta capabilities for
cryoEM refinement and assessment
- Rosetta is a very large and successful
suite of programs from the David Baker lab (University of Washington)
for modeling/building/predicting protein and protein-complex structures,
licensed free of charge to academic and government labs
- Rosetta itself does not have a GUI interface; my impression is that
its users predominantly rely on Pymol for visualization
and most third-party GUI interfaces to it are Pymol-based
- StarMap developers: Frank DiMaio is at University of Washington, whereas
Thomas Marlovits is affiliated with various institutions in Germany and
Austria; both have cited Chimera and ChimeraX many times
- StarMap is open source
(BSD 2-Clause License) but installation may require
IT expertise, especially if for use on a cluster
[back to paper list]
StarMap/Rosetta pipeline
- requires initial structure + cryoEM map (pref. two half-maps); *=optional
- refinement can be done in a day for many systems, whereas the
assessments take just minutes
Refinement
- mainly targeted at medium-to-low resolution cryoEM data (~3.0-4.5 Å)
- all-atom refinement with local density-guided optimization
and real-space B-factor calculation
- optional application of cyclic, dihedral, or helical symmetry
- uses efficient Monte Carlo sampling with the Rosetta energy function,
“...allowing the realistic modeling of flexible stretches of residues
(loops) in lieu of strong density...
the algorithm can equally well handle both fine local optimizations
and large domain motions... small-molecule ligands could be easily modeled”
- they contend that MD-based methods like ISOLDE and Namdinator are more
time-intensive and have a smaller radius of convergence
(are more prone to trapping in local minima). However,
[StarMap] “ ...can be coupled with ISOLDE, an excellent
interactive modeling tool, thus turning ChimeraX into a one-stop shop
for cryo-EM modeling and refinement.”
- streamlined process with not too many user-adjustable parameters in the GUI
(small subset of what's in Rosetta, but advanced users can customize the GUI)
- does not require programming skills; will generate scripts from GUI
settings, allows viewing/editing them before they are executed
- a large multicomponent system can be refined in one run
- different models can be computed in parallel on multiple processors
StarMap GUI Overview
- *Symm: check the symmetry of the input PDB file,
generate symmetry-definition files for refinement
- Rosetta: select refinement strategy, supply map(s) and model
- *Advanced: set symmetry options and constraints
- *User: provide user-defined Rosetta XML scripts,
set replacement variables to customize StarMap templates
- Save / Run: view the generated ChimeraX, Rosetta, and shell scripts,
edit as needed, save the versions that will be executed;
run the refinement locally or on a cluster
- Analysis: rank models, compute per-residue Z-scores
- Apix: pixel-size calibration, B-factor refinement
- Log: diagnostic messages
see intro video (5 min)
Run Locally or on Cluster
Assessment
- a. The Z-score was computed to assess the fit of PDB 6LFO
into its cryo-EM map (EMDB 0879).
In the inset, the map is colored by residue Z-score and
the detergent micelle is gray.
In the main image, the model ribbons are colored by
residue Z-score, with the map silhouette in gray.
The interleukin-8 chain, which exhibits poor Z-scores, is enclosed
in a box.
- b. Interleukin-8 coloring by Z-score components
(overall fit, weight 0.45; fit compared to neighboring residues, 0.05;
bond geometry, 0.15; backbone conformation 0.35),
suggesting that the stereochemistry is fine
but the experimental density is insufficient to support the model.
- As in the example above, assessment is independent of the refinement
and can be run on pre-existing structures with maps available.
Limitations
- best suited for cryoEM; X-ray data can be used, but with additional steps
such as conversion to a real-space map (2Fo-Fc) with Phenix or Coot, and
the model still needs to be refined in reciprocal space with Phenix or Refmac
- requires installing Rosetta, of course, and for parallel execution,
a message-passing interface (MPI) and making a Rosetta MPI executable
since they're not available prefab
- relies on reasonable placement/conformation of initial model
- mostly tested on proteins, and the Z-score is only computed for
protein residues
- although they can be modeled by Rosetta, glycosylations, waters,
and covalently modified residues are not yet handled by StarMap
- monatomic ions are not refined
- parameters for ligands not in the PDB Chemical Component Dictionary
need to be defined separately
- Rosetta does not yet fully support PDBx/mmCIF format