CTFFIND 4

<h1>CTFFIND 4</h1>
October 2014
<h2>What's new?</h2>
<table>
	<tr>
<td>
<ul>
	<li>Much faster (approximately 10x)
	<li>Backward compatibility with CTFFIND 3 by giving <code>--old-school-input</code>. Should not break existing scripts
	<li>Completely rewritten, in object-oriented Fortran
	<li>New interactive user interface with friendly help
	<li>Improved output diagonstics, including an estimate of the maximum resolution at which Thon rings were detected
	<li>Outputs 1D profiles which accurately account for astigmatism
	<li>Accepts stack of micrographs as input
	<li>Accepts stack of movie frames as input
</ul>
</td>
<td>
<img src="ctffind_diag_screenshot.png" width="256px"/>
</td>
</tr>
</table>

<h2>Why is it faster?</h2>
Mainly because the astigmatism azimuth angle is estimated separately from defocus parameters, making the search & refinement 2- rather than 3-dimensional.
Also, many parts of the code were optimized for speed.

<h2>Does it do exactly the same thing as CTFFIND 3?</h2>
No, and it doesn't give exactly the same results, but the results should be very, very similar. Below are some of main differences in processing between
CTFFIND 3 and CTFFIND 4:
<ul>
	<li>The astigmatism angle is estimated separately from the defocus parameters
	<li>Rather than estimating the micrograph power spectrum by averaging spectra from many tiles, a single large Fourier transform of the whole micrograph is computed and then downsampled for CTF fitting.
	<li>All areas of the micrograph are used for power spectrum estimation
	<li>Micrograph edges are "tapered" on input, a la MRC.
	<li>Background subtraction was modified so that it completely ignores frequencies below the minimum frequency used for fitting.
</ul>

Other aspects of the algorithm have not changed from CTFFIND 3 to CTFFIND 4.

<h2>Should I switch from CTFFIND 3 to CTFFIND 4?</h2>
If you are working on micrographs which were collected on photographic film and scanned in, you should probably stick with CTFFIND 3, because micrograph features such as film labels or scratches may throw CTFFIND 4 off, whereas CTFFIND 3 should ignore them. Otherwise, if you are working on images from CCDs or
direct detectors, you should switch to CTFFIND 4: it is significantly faster and may give better estimates of the power spectrum because it will not discard any part of the micrograph.

<h2>Should I use CTFFIND on unaligned movie frames, or on aligned and averaged frames?</h2>
If you think there may be significant errors in your alignment of the movie frames, it may be beneficial to give the unaligned frames as input to CTFFIND 4. If however you are confident in your frame alignment, giving the aligned average as input should give better results and will run much faster. In fact, using CTFFIND both ways for a few test micrographs may give you an indication of whether you had significant alignment errors.

<h2>What exactly is plotted by the ctffind_plot_results.sh script?</h2>
Here is an example of the output from the script:
<img src="notes-140710_ctffind_example_plot.png" />
<h3>Experimental power spectrum - green line</h3>
The green line is a 1D average of the experimental power spectrum obtained from the micrograph. It is computed by taking into account astigmatism. Also, it is normalised to optimise visualisation in the plot:
<ul>
	<li>The minima of the oscillations are set to 0.0
	<li>Values are normalised such that the second peak of the power spectrum (in this case at around 0.06) is at 0.95
	<li>For spatial frequencies at which the power spectrum maxima are less than 0.1, values are normalised such that the values at estimated maxima positions are 0.1. This means that spatial frequencies at which no CTF rings were present in the power spectrum or at which the CTF estimation was wrong will end
	up with very noisy-looking 1D spectra.
</ul>
One consequence of all these normalisation steps is that you should not use this 1D profile to estimate anything quantitative about, say, envelope functions of the microscope.

<h3>CTF fit - orange line</h3>
The orange line is a 1D average of the 2D theoretical CTF which was fit to the power spectrum. The 1D averaging is done taking astigmatism into account.

<h3>Quality of fit - blue line</h3>
The blue line plots a measure of the quality of fit between the theoretical CTF (orange) and the experimental power spectrum (green). The quality of fit is calculated once per "cycle" of the squared CTF - the interval from one maximum to the next. This estimate is computed as a normalised cross-correlation between the 1D profile of the experimental power spectrum and the 1D profile of the fit CTF.


<h2>How does ctffind estimate the highest spacing to which CTF rings were fit successfully?</h2>
This value is in the 6th column of the summary output .txt file. It is estimated as the last spatial frequency at which the quality of fit measure is still above 0.2. This threshold was chosen heuristically. Please take this estimate with a pinch of salt.

<h2>OpenMP threading</h2>
On most systems, the program will use as many OpenMP threads as are available. You can control how many threads are available using standard OpenMP means,
such as setting the environment variable OMP_NUM_THREADS. In addition, you can use the command-line option --omp-num-threads=N or --omp-num-threads N to
limit the program to N threads maximum.

<h2>Author</h2>
CTFFIND 4 was written by <a href="http://grigoriefflab.janelia.org/arohou">Alexis Rohou</a>. It is a rewrite of CTFFIND 3, written by <a href="http://grigoriefflab.janelia.org/ngrigorieff">Nikolaus Grigorieff</a>.

<h2>Known issues</h2>
<ul>
	<li>Only MRC files are supported
</ul>

<h2>Bug reports, questions</h2>
Please direct requests for help and bug reports to the Grigorieff lab forum: <a href="http://grigoriefflab.janelia.org/ctf/forum">http://grigoriefflab.janelia.org/ctf/forum</a>