path: root/doc/man/indexamajig.1

.\"
.\" indexamajig man page
.\"
.\" Copyright © 2012 Thomas White <taw@physics.org>
.\"
.\" Part of CrystFEL - crystallography with a FEL
.\"

.TH INDEXAMAJIG 1
.SH NAME
indexamajig \- bulk indexing and data reduction program
.SH SYNOPSIS
.PP
.BR indexamajig
\fB-i\fR \fIfilename\fR \fB-o\fR \fIoutput.stream\fR \fB-g\fR \fIdetector.geom\fR \fB-b\fR \fIbeamline.beam\fR \fB--peaks=\fR\fImethod\fR \fB--indexing=\fR\fImethod\fR \fB--cell-reduction=\fR\fImethod\fR
[\fBoptions\fR] \fB...\fR
.PP
\fBindexamajig --help\fR

.SH DESCRIPTION

indexamajig takes as input a list of diffraction image files, currently in HDF5 format.  For each image, it attempts to find peaks and then index the pattern.  If successful, it will measure the intensities of the peaks at Bragg locations and produce a list of integrated peaks and intensities (and so on) in an output text file known as a "stream".

For minimal basic use, you need to provide the list of diffraction patterns, the method which will be used to index, a file describing the geometry of the detector, a PDB file which contains the unit cell which will be used for the indexing, and that you'd like the program to output a list of intensities for each successfully indexed pattern.  Here is what the minimal use might look like on the command line:

.IP \fBindexamajig\fR
.PD
-i mypatterns.lst -j 10 -g mygeometry.geom --indexing=mosflm,dirax --peaks=hdf5 --cell-reduction=reduce -b myxfel.beam -o test.stream -p mycell.pdb --record=integrated

.PP
More typical use includes all the above, but might also include a noise or
common mode filter (--filter-noise or --filter-cm respectively) if detector
noise causes problems for the peak detection.  The HDF5 files might be in some
folder a long way from the current directory, so you might want to specify a
full pathname to be added in front of each filename.  You'll probably want to
run more than one indexing job at a time (-j <n>).

You can include a table of saturation values for in the HDF5 file, if you have
a method for estimating the intensities of saturated peaks.  It goes in
/processing/hitfinder/peakinfo_saturated, and should be an n*3 two dimensional
array, where the first two columns contain fast scan and slow scan coordinates
(in that order) and the third contains the value which should belong in a peak
at the given location.  The value will be spread in a small cross centred on
that location.

See \fBman crystfel_geometry\fR for information about how to create a geometry description file.

You can control what information is included in the output stream using
\fB--record\fR=\fIflags\fR.  Possible flags are:

.IP \fBintegrated\fR
.PD
Include a list of reflection intensities, produced by integrating around predicted peak locations.

.IP \fBpeaks\fR
.PD
Include peak locations and intensities from the peak search.

.IP \fBpeaksifindexed\fR
.PD
As 'peaks', but the peak information will be written only if the pattern could be indexed.

.IP \fBpeaksifnotindexed\fR
.PD
As 'peaks', but the peak information will be written only if the pattern could \fInot\fR be indexed.

.PP
The default is \fB--record=integrated\fR.

.PP
If you just want the integrated intensities of indexed peaks, use \fB--record=integrated\fR.  If you just want to check that the peak detection is working, used \fB--record=peaks\fR.  If you want the integrated peaks for the indexable patterns, but also want to check the peak detection for the patterns
which could not be indexed, you might use \fB--record=integrated,peaksifnotindexed\fR and then use \fBcheck-peak-detection\fR from the scripts folder to visualise the results of the peak detection.

.SH PEAK DETECTION

You can control the peak detection on the command line.  Firstly, you can choose the peak detection method using \fB--peaks=\fR\fImethod\fR.  Currently, two values for "method" are available.  \fB--peaks=hdf5\fR will take the peak locations from the HDF5 file.  It expects a two dimensional array at where size in the first dimension is the number of peaks and the size in the second dimension is three.  The first two columns contain the fast scan and slow scan coordinates, the third contains the intensity.  However, the intensity will be ignored since the pattern will always be re-integrated using the unit cell provided by the indexer on the basis of the peaks.  You can tell indexamajig where to find this table inside each HDF5 file using \fB--hdf5-peaks=\fR\fIpath\fR.

If you use \fB--peaks=zaef\fR, indexamajig will use a simple gradient search after Zaefferer (2000).  You can control the overall threshold and minimum gradient for finding a peak using \fB--threshold\fR and \fB--min-gradient\fR.  Both of these have arbitrary units matching the pixel values in the data.

A minimum peak separation can also be provided in the geometry description file
(see \fBman crystfel_geometry\fR for details).  This number serves two purposes.  Firstly, it is the maximum distance allowed between the peak summit and the foot point (where the gradient exceeds the minimum gradient).  Secondly, it is the minimum distance allowed between one peak and another, before the later peak will be rejected.

You can suppress peak detection altogether for a panel in the geometry file by specifying the "no_index" value for the panel as non-zero.


.SH INDEXING METHODS

You can choose between a variety of indexing methods.  You can choose more than one method, in which case each method will be tried in turn until the later cell reduction step says that the cell is a "hit".  Choose from:

.IP \fBnone\fR
.PD
No indexing, peak detection only.

.IP \fBdirax\fR
.PD
DirAx will be executed.  DirAx must be installed and in your PATH for this to work.  Test by running \fBdirax\fR on the command line immediately before running \fBindexamajig\fR - you should see a welcome message.  If not, refer to the installation instructions for DirAx.

.IP \fBmosflm\fR
.PD
MOSFLM will be executed.  MOSFLM must be installed and in your PATH for this to work.  Test by running \fBipmosflm\fR on the command line immediately before running \fBindexamajig\fR - you should see a welcome message.  If not, refer to the installation instructions for CCP4.

.IP \fBreax\fR
.PD
An \fIexperimental\fR DPS-style algorithm will be used, which searches for a lattice with cell parameters close to those contained in the CRYST1 line of the PDB file given with \fB-p\fR or \fB--pdb\fR.  If you use this option, you should also use \fB--cell-reduction=none\fR.

.PP
The default is \fB--indexing=none\fR.


.SH CELL REDUCTION
Neither MOSFLM nor DirAx are currently able to simply search for the orientation of a known unit cell.  Instead, they determine the unit cell parameters ab initio.  To check if the parameters match the known unit cell, which you provide with \fB-p\fR \fIunitcell.pdb\fR or \fB--pdb=\fR\fIunitcell.pdb\fR.  There is a choice of ways in which this comparison can be performed, which you can choose between using \fB--cell-reduction=\fR\fImethod\fR.  The choices for \fImethod\fR are:

.IP \fBnone\fR
.PD
The raw cell from the autoindexer will be used.  The cell probably won't match the target cell, but it'll still get used.  Use this option to test whether the patterns are basically "indexable" or not, or if you don't know the cell parameters.  In the latter case, you'll need to plot a histogram of the resulting parameters from the output stream to see which are the most popular.

.IP \fBreduce\fR
.PD
Linear combinations of the raw cell will be checked against the target cell.  If at least one candidate is found for each axis of the target cell, the angles will be checked to correspondence.  If a match is found, this cell will be used for further processing.  This option should generate the most matches.

.IP \fBcompare\fR
.PD
The cell will be checked for correspondence after only a simple permutation of the axes.  This is useful when the target cell is subject to reticular twinning, such as if one cell axis length is close to twice another.  With \fB--cell-reduction=reduce\fR there would be a possibility that the axes might be confused in this situation.  This happens for lysozyme (1VDS), so watch out.

.PP
The tolerance for matching with \fBreduce\fR and \fBcompare\fR can be set using \fB--tolerance=\fR\fIa,b,c,angl\fR \- see below for more details.  Cells from these reduction routines are further constrained to be right-handed, but no such constraint will be applied if you use \fB--cell-reduction=none\fR.  Always using a right-handed cell means that the Bijvoet pairs can be told apart.

.PP
If the unit cell is centered (i.e. if the space group begins with I, F, H, A, B, C), you should be careful when using "compare" for the cell reduction, since (for example) DirAx will always find a primitive unit cell, and this cell must be converted to the non-primitive conventional cell from the PDB.

.PP
The default is \fB--cell-reduction=none\fR.

.SH PEAK INTEGRATION
If the pattern could be successfully indexed and the cell reduction gave a
positive match, peaks will be predicted in the pattern and their intensities
measured.  The peak integration relies on knowing an accurate radius to
integrate the peak within, and that the annulus used to estimate the background
level is not so big that that it covers neighbouring peaks.  However,
indexamajig cannot (yet) determine these values for you.  You need to specify
them using the \fB--int-radius\fR option (see below).
.PP
To determine appropriate values, index some patterns with the default values and
view the results using \fBcheck-near-bragg\fR (in the scripts folder).  Set the
binning in \fBhdfsee\fR to 1, and adjust the ring radius until none of the rings
overlap for any of the patterns.  This ring radius is the outer radius to use.
Then reduce the radius until the circles match the sizes of the peaks as
closely as possible.  This value is the inner radius.  The middle radius should
be between the two, ideally between two and three pixels smaller than the outer
radius.
.PP
If it's difficult to do this without setting the middle radius to the
same value as the inner radius, then the peaks are too close together to be
accurately integrated.  Perhaps you got greedy with the resolution and put the
detector too close to the interaction region?  Improved integration algorithms,
designed to handle such difficult cases, are under development.

.SH OPTIONS
.PD 0
.IP "\fB-i\fR \fIfilename\fR"
.IP \fB--input=\fR\fIfilename\fR
.PD
Read the list of images to process from \fIfilename\fR.  The default is \fB--input=-\fR, which means to read from stdin.

.PD 0
.IP "\fB-o\fR \fIfilename\fR"
.IP \fB--output=\fR\fIfilename\fR
.PD
Write the output data stream to \fIfilename\fR.  The default is \fB--output=-\fR, which means to write to stdout.

.PD 0
.IP \fB--peaks=\fR\fImethod\fR
.PD
Find peaks in the images using \fImethod\fR.  See the second titled \fBPEAK DETECTION\fB (above) for more information.

.PD 0
.IP \fB--indexing=\fR\fImethod\fR
.PD
Index the patterns using \fImethod\fR.  See the section titled \fBINDEXING METHODS\fR (above) for more information.

.PD 0
.IP \fB--cell-reduction=\fR\fImethod\fR
.PD
Use \fImethod\fR for unit cell reduction.  See the section titled \fBCELL REDUCTION\fR (above) for more information.

.PD 0
.IP "\fB-g\fR \fIfilename\fR"
.IP \fB--geometry=\fR\fIfilename\fR
.PD
Read the detector geometry description from \fIfilename\fR.  See \fBman crystfel_geometry\fR for more information.

.PD 0
.IP "\fB-b\fR \fIfilename\fR"
.IP \fB--beam=\fR\fIfilename\fR
.PD
Read the beam description from \fIfilename\fR.  See \fBman crystfel_geometry\fR for more information.

.PD 0
.IP "\fB-p\fR \fIfilename\fR"
.IP \fB--pdb=\fR\fIfilename\fR
.PD
Read the unit cell for comparison from the CRYST1 line of the PDB file called \fIfilename\fR.

.PD 0
.IP "\fB-e\fR \fIpath\fR"
.IP \fB--image=\fR\fIpath\fR
.PD
Get the image data to display from \fIpath\fR inside the HDF5 file.  For example: \fI/data/rawdata\fR.  If this is not specified, the default behaviour is to use the first two-dimensional dataset with both dimensions greater than 64.

.PD 0
.IP \fB--int-radius=\fR\fIinner,middle,outer\fR
.PD
Set the inner, middle and outer radii for three-ring integration.  See the
section about \fBPEAK INTEGRATION\fR, above, for details of how to determine
these.  The defaults are probably not appropriate for your situation.
.PD
The default is \fB--int-radius=4,5,7\fR.


.PD 0
.IP \fB--basename\fR
.PD
Remove the directory parts of the filenames taken from the input file.  If \fB--prefix\fR or \fB-x\fR is also given, the directory parts of the filename will be removed \fIbefore\fR adding the prefix.

.PD 0
.IP "\fB-x\fR \fIprefix\fR"
.IP \fB--prefix=\fR\fIprefix\fR
.PD
Prefix the filenames from the input file with \fIprefix\fR.  If \fB--basename\fR is also given, the filenames will be prefixed \fIafter\fR removing the directory parts of the filenames.

.PD 0
.IP \fB--hdf5-peaks=\fR\fIpath\fR
.PD
When using \fB--peaks=hdf5\fR, read the peak locations from a table in the HDF5 file located at \fIpath\fR.

.PD 0
.IP \fB--tolerance=\fR\fItol\fR
.PD
Set the tolerances for unit cell comparison.  \fItol\fR takes the form \fIa\fR,\fIb\fR,\fIc\fR,\fIang\fR.  \fIa\fR, \fIb\fR and \fIc\fR are the tolerances, in percent, for the respective direct space axes when using \fBcompare\fR or \fBcompare_ab\fR for cell reduction (see above).  \fIang\fR is the tolerance in degrees for the angles.  They represent the respective \fIreciprocal\fR space parameters when using \fB--cell-reduction=reduce\fR.
.PD
The default is \fB--tolerance=5,5,5,1.5\fR.

.PD 0
.IP \fB--filter-cm\fR
.PD
Attempt to subtract common-mode noise from the image.  The filtered image will be used for the final integration of the peaks (in contrast to \fB--filter-noise\fR.  It is usually better to do a careful job of cleaning the images up before using indexamajig, so this option should not normally be used.

.PD 0
.IP \fB--filter-noise\fR
.PD
Apply a noise filter to the image with checks 3x3 squares of pixels and sets all of them to zero if any of the nine pixels have a negative value.  This filter may help with peak detection under certain circumstances, and the \fIunfiltered\fR image will be used for the final integration of the peaks.   It is usually better to do a careful job of cleaning the images up before using indexamajig, so this option should not normally be used.

.PD 0
.IP \fB--no-sat-corr\fR
.PD
This option is here for historical purposes only, to disable a correction which is done if certain extra information is included in the HDF5 file.

.PD 0
.IP \fB--threshold=\fR\fIthres\fR
.PD
Set the overall threshold for peak detection using \fB--peaks=zaef\fR to \fIthres\fR, which has the same units as the detector data.  The default is \fB--threshold=800\fR.

.PD 0
.IP \fB--min-gradient=\fR\fIgrad\fR
.PD
Set the gradient threshold for peak detection using \fB--peaks=zaef\fR to \fIgrad\fR, which units of "detector units per pixel".  The default is \fB--min-gradient=100000\fR.

.PD 0
.IP \fB--min-snr=\fR\fIsnr\fR
.PD
Set the minimum I/sigma(I) for peak detection when using \fB--peaks=zaef\fR.  The default is \fB--min-snr=5\fR.

.PD 0
.IP \fB--min-integration-snr=\fR\fIsnr\fR
.PD
Set the minimum I/sigma(I) for a peak to be integrated successfully.  The default is \fB--min-snr=-infinity\fR, i.e. no peaks are excluded.

.PD 0
.IP \fB--copy-hdf5-field=\fR\fIpath\fR
.PD
Copy the information from \fIpath\fR in the HDF5 file into the output stream.  The information must be a single scalar value.  This option is sometimes useful to allow data to be separated after indexing according to some condition such the presence of an optical pump pulse.  You can give this option as many times as you need to copy multiple bits of information.

.PD 0
.IP \fB--verbose\fR
.PD
Be more verbose about indexing.

.PD 0
.IP "\fB-j\fR \fIn\fR"
.PD
Run \fIn\fR analyses in parallel.  Default: 1.

.PD 0
.IP \fB--no-check-prefix\fR
.PD
Don't attempt to correct the prefix (see \fB--prefix\fR) if it doesn't look correct.

.PD 0
.IP \fB--closer-peak\fR
.PD
If you use this option, indexamajig will integrate around the location of a detected peak instead of the predicted peak location if one is found close to the predicted position, within ten pixels.  \fBDon't use this option\fR, because
there is currently no way to set the definition of 'nearby' to be appropriate
for your data.

.PD 0
.IP \fB--no-closer-peak\fR
.PD
This is the opposite of \fB--closer-peak\fR, and is provided for compatibility
with old scripts because this option selects the behaviour which is now the
default.

.PD 0
.IP \fB--insane\fR
.PD
Normally, indexamajig will check that the projected reciprocal space positions of peaks found by the peak detection are close to reciprocal lattice points.  This option disables this check, and normally is \fInot\fR a good idea.

.PD 0
.IP \fB--no-bg-sub\fR
.PD
Don't subtract local background estimates from integrated intensities.  This is almost never useful, but might help to detect problems from troublesome background noise.

.PD 0
.IP \fB--cpus=\fR\fIn\fR
.IP \fB--cpugroup=\fR\fIn\fR
.IP \fB--cpuoffset=\fR\fIn\fR
.PD
See the section below about \fBTUNING CPU AFFINITIES FOR NUMA HARDWARE\fR.  Note in particular that \fB--cpus\fR is not the same as \fB-j\fR.



.SH TUNING CPU AFFINITIES FOR NUMA HARDWARE

If you are running indexamajig on a NUMA (non-uniform memory architecture)
machine, a performance gain can sometimes be made by preventing the kernel from
allowing a process or thread to run on a CPU which is distant from the one on
which it started.  Distance, in this context, might mean that the CPU is able to
access all the memory visible to the original CPU, but perhaps only relatively
slowly via a cable link.  In many cases a group of CPUs will have direct access
to a certain region of memory, and so the process may be scheduled on any CPU in
that group without any penalty.  However, scheduling the process to any CPU
outside the group may be slow.  When running under Linux, indexamajig is able to
avoid such sub-optimal process scheduling by setting CPU affinities for its
threads.  The CPU affinities are also inherited by subprocesses (e.g. MOSFLM or
DirAx).

To do this usefully, you need to give indexamajig some information about your
hardware's architecture.  Specify the size of the CPU groups using
"--cpugroup=<n>".  You also need to specify the overall number of CPUs, so that
the program knows when to 'wrap around'.  Using "--cpuoffset=<n>", where "n" is
a group number (not a CPU number), allows you to manually skip a few CPUs, which
may be useful if you do not want to use all the available CPUs but want to avoid
running all your jobs on the same ones.

Note that specifying the above options is NOT the same thing as giving the
number of analyses to run in parallel (the 'number of threads'), which is done
with "-j <n>".  The CPU tuning options provide information to indexamajig about
how to set the CPU affinities for its threads, but it does not specify how many
threads to use.

Example: 72-core Altix UV 100 machine at the author's institution

This machine consists of six blades, each containing two 6-core CPUs and some
local memory.  Any CPU on any blade can access the memory on any other blade,
but the access will be slow compared to accessing memory on the same blade.
When running two instances of indexamajig, a sensible choice of parameters might
be:

.PP
Instance 1: --cpus=72 --cpugroup=12 --cpuoffset=0 -j 36
.PP
Instance 2: --cpus=72 --cpugroup=12 --cpuoffset=36 -j 36

This would dedicate half of the CPUs to one instance, and the other half to the
other.


.SH A NOTE ABOUT UNIT CELL SETTINGS

At the moment, CrystFEL's core symmetry module only knows about one setting for each unit cell.  You must use the same setting for now, but this will be improved in future versions.  The cell settings are the standard ones from the International Tables (2006).  That means, for example, that for orthorhombic cells in point group mm2 the twofold axis should be along "c", i.e. no mirror perpendicular to "c".  For tetragonal cells and hexagonal lattices, the unique axis should be "c" as usual.  For monoclinic cells, the unique axis must be "c".


.SH BUGS
Don't run more than one indexamajig jobs simultaneously in the same working
directory - they'll overwrite each other's DirAx or MOSFLM files, causing subtle
problems which can't easily be detected.
.PP
ReAx indexing is experimental.  It works very nicely for some people, and crashes for others.  In a future version, it will be improved and fully supported.

.SH AUTHOR
This page was written by Thomas White.

.SH REPORTING BUGS
Report bugs to <taw@physics.org>, or visit <http://www.desy.de/~twhite/crystfel>.

.SH COPYRIGHT AND DISCLAIMER
Copyright © 2012 Deutsches Elektronen-Synchrotron DESY, a research centre of the Helmholtz Association.
.P
indexamajig, and this manual, are part of CrystFEL.
.P
CrystFEL is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
.P
CrystFEL is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
.P
You should have received a copy of the GNU General Public License along with CrystFEL.  If not, see <http://www.gnu.org/licenses/>.

.SH SEE ALSO
.BR crystfel (7),
.BR crystfel_geometry (5),
.BR process_hkl (1),
.BR partialator (1)