Joachim Frank

1
Topic 3

• Joachim Frank

2
Generation of initial models, and fitting methods

• Initial models usually generated by hand, using O
  as stereo modeling environment
• Fitting methods
• A) single component rigid-body search using
  locally normalized CC
• B) multiple component flexible fitting
• 1) real-space refinement
• 2) NMFF (normal mode-based flexible fitting)
• 3) (when applicable) MD simulations with CC
  against observed density map

J. Frank
3
Example NMFF of protein-conducting channel
SecYEG channel bound to the translating
ribosome One model (front to front arrangement
of SecYEG heterotrimers) explains densities for
both inactive and active channels. Application
of NMFF (normal mode-based flexible fitting).
(see next slide)
J. Frank
Mitra et al., Nature, in press
4
J. Frank
Mitra et al., Nature, in press
5
Quality of fitted model

• Subjective criteria
• (i) the extent to which the model explains all
  observed density is any density left
  unexplained?
• (ii) do any parts of the atomic model stick
  out of the observed density?
• (iii) for unreconciled multi-component docking
  are different parts of the model in steric
  conflict? Can the conflict be removed by
  postulating conformational changes?
• Objective/quantitative criteria
• Real-space refinement and NMFF return measures
  for goodness of fit, from which the final
  quality can be gauged
• Correlation coefficient, R-factor, and Error
  Residual (see examples in next slide)

J. Frank
6
Goodness of fit by real-space refinement, as a
function of the number of pieces used In the
flexible fitting of the ribosome.
Goodness of fit criteria, and their sensitivity
to shift of the model away from the correct
position.
J. Frank
7
Validation requirements

• 1) Reproducibility of the fitting Apply fitting
  procedure to two independent (not halfset)
  reconstructions, and compare the results using
  rmsd.
• 2) Reproducibility of conformational changes
  deduced When real-space refinement is used on
  two density maps A and B, to obtain two models
  from which conformational changes are to be
  deduced, then we need a control reconstruction,
  A1, to which the same procedure is applied.
• Rmsd between the relevant components of the
  two structural models should be significantly
  larger (factor of 3) than the rmsd between A and
  A1.

J. Frank
8
Archiving of fitted models

• Internal deposition (Archive Database)
• Public deposition in the pdb, in conjunction with
  publications
• We try to deposit the cryo-EM maps simultaneously
  in the 3DEM data base. (Ribosome complexes
  deposited by our group constitute the largest
  portion of the existing 3DEM data base.)

J. Frank