Flex-EM   image

Fitting and refinement of atomic structures guided by cryoEM density.     

Please cite: M. Topf, K. Lasker, B. Webb, H. Wolfson, W. Chiu, A. Sali. Structure 16, 295-307, 2008 PDF

Other citations: A.P. Joseph, S. Malhotra, T. Burnley, C. Wood, D.K. Clare, M. Winn, M. Topf. Methods 100, 42-49, 2016 PDF

The method includes a rigid fitting stage followed by a refinement stage. Rigid fitting can be performed with Mod-EM or any other rigid fitting method. The refinement stage starts with the components rigidly fitted in the approximate positions in the map. Two methods are available: conjugate gradients minimization (CG) and simulated annealing molecular dynamics (MD). To use it:

A. Download the following scripts:

B. Edit the following files:

flex-em.py (below INPUT PARAMETERS):
1. Define the mode of optimization (optimization='MD' or optimization='CG'). 
2. Set the input parameters of the atomic structure (PDB format) that you want to fit and the EM map (MRC or XPLOR format).
3. Specify the path of your work directory.
4. For CG optimization: specify the number of CG runs (eg, 
num_of_runs=2) and the number of the initial directory in which you want the runs to be found (eg, init_dir=1 will produce the results in the directories 1_cg, 2_cg).
5. For MD optimization: specify the number of MD iterations (eg, 
num_of_iter=4) and the number of the directory in which you want the runs to be found (eg, init_dir=1 will produce the results in the directory 1_md).

Important Notes:
- If your origin is in pixels (e.g. from Chimera) please multiply by -1*voxel size.
- If you have chain breaks in your structure keep the TER line in your PDB input file (otherwise the program will connect your chains).
- If you know the approximate initial positions of your components running the MD refinement is recommended.
- If you are not sure about the approximate positions of the components, it is recommended:
  (a) First run the CG minimization multiple times (eg, 
num_of_runs = 20);
  (b) Cluster the different "fits";
  (c) Once you have selected your ensemble of representative "rigid fits" you can start an MD refinement for each of them (using each as an initial file for the MD optimization (eg, 
input_pdb_file='final1_cg.pdb' ).
- For high-resolution refinement (better than ~4.5 Å) change the cap_shift in  flex-em.py . SSE restrained: cap_shift=0.15, all-atom refinement: cap_shift=0.10.

Define the rigid bodies using residue numbers. The format of this file is the following:
- Comment lines begin with '#'  (eg, describing what kind of a rigid body : 
- Other lines: each line describes one rigid body by specifying the initial and final residue of each of the segments in that rigid body
 '2 6 28 30' means that residues 2-6 and 28-30 will be included in the same rigid body).
- For multi-chain models: each line describes one rigid body by specifying the initial and final residue of each of the segments, along with their chain IDs
 '2:A 6:A 28:B 30:B' means that residues 2-6 from chain A and 28-30 from chain B will be included in the same rigid body).
- For including ligands and other hetero-atoms (HETATM), the molecule number and chain ID can be added to the rigid body line in the same format (but it should not form a separate rigid body).
 '100:A 100:A' means that the HETATM 100 from chain A will be considered as a rigid body).
- Only in CG mode: if you add 'nr' after the last segment (eg,
'2 6 28 30 nr') the rigid body will not be randomized prior to the running of the optimization.

Important Notes:
- For a better description of your rigid bodies please use RIBFIND (see: Pandurangan & Topf, J Struct Biol 2011Pandurangna_JSB_2011 ). The program will automatically produce your rigid.txt file.

C. Run the program (using MODELLER-9v2 or later):
mod9v9 flex-em.py > flex-em.log 

The refined structures will be found in the specified directory as described above.

Flex-EM can be applied to both experimental atomic structures and homology models. tutorial: flex_em_tutorial.pdf.


Flex-EM benchamrk is available for download here.

- To minimize bias, the benchmark density maps were not produced with Mod-EM, but with pdb2vol in SITUS (Wriggers, et al. 1999), which uses a different Gaussian smoothing technique.
- Cα RMSDs were calculated for the benchmark using the superpose method in MODELLER-9v2.

For questions, please contact Maya Topf at m.topf@cryst.bbk.ac.uk