mode 3 search range

Submitted by mbycliaw on Sun, 2011-05-08 21:02

Forums

Frealign

Hi,

If we have a icosahedral virus, under mode=3,
setting search angle, does program still search whole space or only search asymmetric unit (half triangle)?

If still search the whole space, is there a way to provide a list of angles for frealign to search?

Thanks

niko Mon, 2011-05-09 09:27

In mode=3, Frealign simply tests random orientations. They will not be limited to the asymmetric unit. There is no easy way to supply Frealign with a list of specific angles to search over. One possibility would be to make a stack for each particle that simply repeats the particle image N times (where N is the number of orientations you want to search). Then you can enter in the Frealign parameter file for this (artificial) stack the angles you want to search and run Frealign in mode=1. Alternatively, you can modify the Frealign code, of course.

mbycliaw Wed, 2011-05-11 08:21

In the 2011 EMBO J. rotavirus paper, the sentence "TLP images were 2x binned and aligned to previous....through a systematic search at a 1 dgree angular interval (FREALIGN mode 3)". So you mean it actually search the whole space in 1 degree interval. That's a hugh calculation, isnt it? Is there a reason to do that? is it much better than mode 4 with bigger interval?

niko Wed, 2011-05-11 10:08

You are right, it took a while to finish this search. I just realized that I did not answer your question correctly. In Mode=3 Frealign carries out a systematic search on a grid (for example with 1 deg spacing). Therefore, there are no random orientations with Mode=3. The grid does depend on the starting angles provided in the input parameter file, however.

We have a computer cluster that allowed us to parallelize searches like this. It is certainly true that using Frealign is not the most effective way to determine initial angles and shifts. Frealign was primarily designed to carry out refinement. Apart from a systematic search, there are also other search strategies, for example with Mode=4 which allows a grid search with random start angles. Please read the 2007 paper on Frealign for more info. It is usually a good idea to test the chosen search strategy on a few particles and set FMATCH=T to generate matching projections on output. Then check the matching projections and make sure Frealign found the correct angles. Once the search has been optimized, FMATCH can be set back to F to save disk space and speed up the calculation.

mbycliaw Tue, 2012-02-14 20:45

I am not clear about mode 3. In mode 3 the parameter DANG define spacing angle. But there is another parameter ITMAX define the repeat, said 5. Does it mean it only carry out 5x5x5 grid on three angular space search? (which means it is not a whole space search and that also explain it depend on starting angle, right?)
or it still carry out full space search, then what does ITMAX mean here?

For mode 3 with 1 degree spacing we only need to run it once. For mode 4 with large spacing angle and large repeat, said 200, do we need to run it more than one time?

This search mode should carry out in low resolution and refine graduately to high resolution, right? It seems possible that the orientation of some particles might be trapped in a local minimum, right? Is there a way to overcome it?

niko Tue, 2012-02-14 21:20

The different search strategies are described in Grigorieff (2007). Mode = 3 will run a systematic search once, followed by refinement of the orientations with the best phase residuals (internally: correlation coefficients). ITMAX will not have any effect on this mode. In mode = 5, Frealign will perform systematic grid searches with randomly selected starting points. Therefore, you will have differently placed grids. But the grids are covering the entire angular space. You can define very coarse grids and repeat the search with random starting points many times. Or you could have a very tight grid and do it only once. You have to try what works best for your data. Global searches should not get stuck in local optima.