Refinement in FREALIGN after Spider projection matching

Submitted by pgrob on Mon, 2010-01-25 19:16

Forums

Frealign

We are getting our initial particle orientations and shifts from Spider projection matching (CTF parameters have been determined using CTFFIND3). We then want to do the final refinement and CTF correction in FREALIGN. However we usually end up with the same resolution after the refinement in FREALIGN as before. We seem to be stuck in the 30-20A resolution range. Is there a particular way to go about the FREALIGN refinement (incrementally changing the resolution of the refinement, etc)? Should we get our initial alignment parameters a different way? Or are we just limited by our data? Are there some FREALIGN refinement modes that would improve substantially the initial alignment (CTF parameter refinement, use of solvent flattening, negative B-factor, phase residual threshold, ...), or allow us to escape local phase residual minima?
Any suggestion would help.
Thanks!

niko Mon, 2010-01-25 20:49

The answer to these questions can be manifold, of course. The resolution could be limited by the data. However, there are a few things one can try:

  1. You could set the RBFACT parameter to a negative number (for example -500 or -1000) for 2 – 3 refinement cycles. This will boost the high-resolution terms during refinement and may lead to some changes in the alignment parameters because of the increased noise from the high-resolution terms. The parameter changes may get you out of local minima. A negative RBFACT will also lead to over-refinement and an apparent improvement of the FSC curve (see Stewart & Grigorieff, 2004). So it is important to set RBFACT back to zero after finishing the 2 or 3 cycles and perform another 10 cycles or so to reduce over-refinement effects.
  2. You can run Frealign in Mode 2 to check if you can find better alignment parameters for your particles. If the parameters are already good they should not change much (check the parameter shifts file to see changes). If many particles show parameter changes it may mean that your initial parameters were inaccurate or that the particles exhibit heterogeneity. The latter could mean that your resolution is limited by the data.
  3. Although Frealign uses a weighted correlation coefficient for the refinement that should downweight high-resolution terms if your reference structure does not have high-resolution signal, it probably helps to limit the resolution more or less to the resolution of your reference structure (parameter RMAX2). RMAX1 should be kept at 100 or 200 Angstrom to include the strong low-resolution terms in the refinement. These terms help in the alignment.
  4. At 20 – 30 Angstrom resolution it will not make sense to do any CTF or magnification refinement. The refinement probably will not work (not enough signal) and your resolution is not likely to be limited by small defocus or magnification errors. CTF and magnification refinement should probably not be done until you close to reach 5 Angstrom resolution, and you have chunky particles that produce strong contrast.
  5. You can try to improve the reconstruction by including only the very best particles. You should play with the threshold parameter THRESH and monitor the FSC curve (just calculate a few reconstructions using different THRESH values, without additional refinement). You might find an optimal number of particles to include.
  6. You can perform multi-reference refinement using the developmental version of Frealign (available from the author upon request) and see if the use of several references will improve the resolution. This might be the case if your particle exhibits heterogeneity. To test this, it might be best to start with only 2 or 3 references and see if the multi-reference refinement yields plausible structures (with differences that you can explain with the biochemistry or function of your particle). Using several references will require more data, of course, as the data set is divided amongst several reconstructions.