Boulder Laboratory for 3-Dimensional Electron Microscopy of Cells
SOLVEMATCH(1) SOLVEMATCH(1)
NAME
solvematch - Solve for transformation matching one tomogram to another
SYNOPSIS
solvematch
DESCRIPTION
Solvematch will solve for a 3-dimensional linear transformation
relating the tomogram volumes resulting from tilt series around two
different axes. It can use information from the 3-D coordinates of
fiducials found by Tiltalign, or from IMOD models of corresponding
points in the two volumes, or from both of these sources together.
Current versions of Tiltalign produce fiducial lists with coordinates
corresponding the coordinates in the tomogram, thus providing all of
the information that Solvematch needs to find the transformation. Older
versions of Tiltalign produced coordinates centered around zero; this
situation is discussed below.
Using Fiducial Coordinates
When you use fiducial coordinates, the program needs to know how the
fiducial points correspond between the two tilt series. There are two
different ways that this information can be provided.
If Transferfid was used to produce the seed model for one of the axes
from the fiducial model for the other, then it can be given an option to
produce a file with the coordinates of corresponding fiducial points from
one view in each data set. This file can be provided to Solvematch with
the "-transfer" option and the program will be able to deduce how the
points in the fiducial lists from Tiltalign correspond. This
capability allows you to add or delete fiducial points in either fiducial
model after running Transferfid without having to keep track of how
this affects the correspondence between points. This option requires
that both fiducial models be supplied to Solvematch. Also, it is
essential that Solvematch be informed if the A axis is being matched to the
B axis, with the -atob option.
In the absence of a transfer coordinate file, the program needs lists of
points that correspond between the two data sets. If it is given
a small list of points which do correspond, it can find the rest of the
correspondences and ignore any points from either set that do not have a
mate. There are two restrictions in using this capability. First, one
must identify at least 4 initial correspondences; in fact, having at
least 5 is recommended. Second, if there are fiducials on both surfaces
of the section, this initial list must include at least one from each
surface.
When you enter point numbers, use the numbers listed on the left in
the fiducial coordinate file. These numbers may not correspond to
the contour numbers in the original fiducial model if some contours
were not included in the solution. These numbers do not need to be
sequential; i.e., you can delete points from the coordinate file.
The program will refer to points by these numbers. When you have
both fiducial coordinates and points from matching models, it will
refer to the latter points by the negative of their point number.
When using fiducial coordinates, you also need to indicate to Solvematch
whether the gold fiducials are on one surface or distributed in depth
(typically but not necessarily on two surfaces). If there is gold on
only one surface, you must determine whether the two tomograms are
inverted in Y (Z in flipped orientation); i.e. whether the gold is on
the "top" of the section in one tomogram and the "bottom" in the other even
though they can be superimposed by a rotation. The program needs to know
this because there is no reliable information about this polarity in the
fiducial coordinates. However, such an inversion should be impossible when
data are acquired with a CCD camera.
Using Matching Models
When fiducial points from Tiltalign are not available, you can use
models of corresponding points to register the two volumes. Pick a set
of points that can be localized well in both tomograms and enter them in
the same order into a model for each tomogram. The points can be in the
same or in different objects or contours, as long as they occur in the
same order in each model. The points should be well-distributed in
all dimensions, and they must not all be in one plane. At least 8 points
are recommended. If this is to provide the basis for the final alignment of
the volumes, they should be relatively near the 8 corners of the volume, and
more points should be used for greater accuracy; but if the alignment is to
be refined with correlation, this is not necessary.
Using Old, Relative Fiducial Coordinates
With fiducial coordinates from older versions of Tiltalign, the program
can solve for the 3 by 3 transformation but not for the shifts in X, Y
and Z. There are two ways to get these shifts: by cross-correlation
of one transformed volume with the other, and by giving Solvematch
models of corresponding points in the two tomograms. The shell script
that does correlations, Matchshifts, is generally successful. If it
fails, you have two options: rerun Tiltalign to get absolute
coordinates; or create matching models. If you do this, 3-6 points are
recommended. If fiducials are essentially on one surface, then the
modeled points are needed to find the scaling in Y (Z in flipped
orientation) between the two tomograms. In this case, be sure to
distribute the points in depth (rather than all on one surface) and use
5-6 points so as to provide information on this scaling.
Local Fitting
Sometimes the fit to the fiducial points gives a high maximum residual
error because there is a big nonlinear distortion between the two
volumes. In these cases, fitting to local sets of points can be used to
determine that the fiducials have been properly identified and that the
global fit is as good as it is ever going to be. If the LocalFitting
option is selected, the program will try local fits when the maximum
residual is above the specified limit. It determines the size of the area
that, on average, would contain the specified minimum number of points.
It then sets up local areas of this size, overlapping by 50%, and expands
an area if necessary to contain the minimum number of points. It performs
fits to the points in each area and reports the results of the fits. If
the overall maximum residual is less than 1.5 times the specified limit,
and no more than 5% of the areas have maximum residuals above this limit,
the fit is considered adequate.
If local fitting is being done, then the program will also compute a local
fit to points at the center of the volume so as to determine how far the
global 3D transformation will shift data at the center of the volume out
of alignment. If this shift is large enough, then corrsearch3d may
have trouble starting to find the alignment of the data. Thus, if the
program finds that this shift is bigger than a specified limit, it
recommends initial shifts to be used when running corrsearch3d and may
also recommend that the initial matching volume be made the same size as
the volume being transformed. It will exit with an error so that you can
adjust these parameters before proceeding.
Program Operation
The program will automatically remove "outliers", pairs of points
that are likely to be incorrect because their residual errors are
so extreme relative to the other points. Up to 10% of points may
be removed in this way. If, even after removing such outliers, the
maximum residual in the linear fit is still higher than a value that
you specify, then the program will exit with an error.
The program determines the transformation and computes the deviation
between each actual point in the first tomogram and the transformation of
the corresponding point in the second tomogram. It reports the mean and
S.D. of these deviations, the number of the point with the highest
deviation, and the magnitude of its deviation.
Inputs and Options
Solvematch uses the PIP package for input (see the manual page for pip)
and can also take input interactively to maintain compatibility with old
command files. The following options can be specified either as command
line arguments (with the -) or one per line in a command file or parameter
file (without the -):
-output OR -OutputFile File name
Output file for 3D transformation. This entry may be omitted.
-afiducials OR -AFiducialFile File name
Name of file with coordinates of fiducials from the first tilt series. To
solve for the transformation using matching models only, omit this entry.
-bfiducials OR -BFiducialFile File name
Name of file with coordinates of fiducials from the second tilt series.
This entry is not needed when using matching models only.
-alist OR -ACorrespondenceList List of integer ranges
A list of the points in series 1 for which a corresponding point in series
2 is known with confidence. Ranges may be entered (e.g. 1-5,7,6,12).
This entry and the following one are not needed when a coordinate file
from transferfid is available. The default is that all points correspond
between the two series.
-blist OR -BCorrespondenceList List of integer ranges
A list of the corresponding points in the second series. The same number
of values must be entered in this list as in the preceding list. Again,
the default is that all points correspond between the two series.
-transfer OR -TransferCoordinateFile File name
File of matching fiducial coordinates produced by running transferfid with
the -c option.
-amodel OR -AFiducialModel File name
Name of fiducial model file for first tilt series. This entry is needed
only when a transfer coordinate file is being used.
-bmodel OR -BFiducialModel File name
Name of fiducial model file for second tilt series. This entry is needed
only when a transfer coordinate file is being used.
-use OR -UsePoints List of integer ranges
List of starting points to use from the A axis (the first series taken).
The program will start with these correspondences and add other pairs of
points that appear to correspond.
-atob OR -MatchingAtoB
The first series here (referred to as A in these options) is the B axis,
i.e. the second tilt series taken, and the second series is the A axis.
-xtilts OR -XAxisTilts Two floats
The values of X axis tilt that were used in generating the first and
second tomograms. These angles are needed to make the fiducial
coordinates from Tiltalign correspond to the actual positions of the
particles in the tomograms. The default values are 0.
-angles OR -AngleOffsetsToTilt Two floats
The values of tilt angle offsets that were used when running Tilt to
generate the first and second tomograms. These are different from the
angle offsets used when running Tiltalign, which need not be supplied to
Solvematch. The default values are 0.
-zshifts OR -ZShiftsToTilt Two floats
The values of Z shift that were used when running Tilt to generate the
first and second tomograms. These are different from the Z shifts used
when running Tiltalign, which need not be supplied to Solvematch. The
default values are 0.
-surfaces OR -SurfacesOrUseModels Integer
This entry governs the mode of operation of the program. To obtain the
transformation from fiducial points only, enter:
2 if there are fiducials on two surfaces,
1 if fiducials are on one surface and the tomograms are NOT inverted
relative to each other,
-1 if the tomograms are inverted and there are fiducials on only one
surface,
0 to use model files of corresponding points; with this entry both the
matching model points and fiducial points will be used if the fiducials
are available.
-2 to use only the matching model points and ignore any entries of
fiducial point files.
The default is 2.
-maxresid OR -MaximumResidual Floating point
The limiting value for the maximum residual, so that the program will exit
with an error if this value is exceeded.
-local OR -LocalFitting Integer
Minimum number of points required in each local area if local fitting is
desired. The default is 0, which disables local fits. Non-zero values
must be at least 6.
-center OR -CenterShiftLimit Floating point
Maximum shift between central local fit and global fit. This shift is
computed and tested against this limit only when local fitting occurs,
namely when the -local option is selected and the global maximum residual
is greater than the -maxresid entry. If the shift is greater than the
limit, the program will exit with an error, but if the solution is
adequate there is no need to rerun the program. The default value is 10.
A value of 0 disables the test.
-amatch OR -AMatchingModel File name
Name of model of points from the first tomogram
-bmatch OR -BMatchingModel File name
Name of model of corresponding points from the second tomogram
-atomogram OR -ATomogramOrSizeXYZ Text string
Either the file name or the X, Y and Z dimensions of the first tomogram.
-btomogram OR -BTomogramOrSizeXYZ Text string
Either the file name or the X, Y and Z dimensions of the second tomogram.
-scales OR -ScaleFactors Two floats
Amounts by which the fiducial coordinates need to be scaled to match the
coordinates of the first and second tomograms, respectively. Each factor
should be the binning of the images on which the fiducials were modeled
divided by the binning of the aligned stack. The entry should not be
needed in ordinary circumstances. It overrides the scaling that is
computed from a pixel size in the fiducial file and the pixel spacing from
the tomogram header. The entry can be used as a substitute for providing
the image file name in the -atomogram and -btomogram options.
-aniso OR -AnisotropicLimit Floating point
The maximum percentage difference between the scaling along the three
axes; a percentage difference greater than this limit will cause a warning
to be issued. If the difference is in the Y direction and points were
specified as being on two surfaces, the program will also advise that it
be run with "-surfaces 1" instead. Enter 0 to disable these warnings.
The default is 10%.
-param OR -ParameterFile Parameter file
Read parameter entries as keyword-value pairs from a parameter file.
-help OR -usage
Print help output
-StandardInput
Read parameter entries from standard input.
Old Style Input
If the program is started with no command line arguments, it reverts to
interactive input and is restricted to older modes of operation. The
input takes different forms depending on what data are available. The
inputs to the program when fiducials from Tiltalign are available are:
Name of file with coordinates of fiducials from the first tilt series
Name of file with coordinates of fiducials from the second series
A list of the points in series 1 for which a corresponding point in
series 2 is known with confidence, or / if all points correspond between
the two series.
A list of the corresponding points in the second series, or / if all points
correspond between the two series.
The values of X axis tilt that were used in generating the first and
second tomograms.
The limiting value for the maximum residual, so that the program will
exit with an error if this value is exceeded.
Enter either 0 to determine shifts from model files of corresponding
points, or -1, 1, or 2 to determine just the 3 by 3 transformation
and not the shifts. Enter 2 if there are fiducials on two
surfaces, 1 if fiducials are on one surface and the tomograms are
NOT inverted relative to each other, or -1 if the tomograms are
inverted and there are fiducials on only one surface.
IF you entered 0, make the following four entries:
Either the file name or the X, Y and Z dimensions of the first
tomogram
Name of model of points from the first tomogram
Either the file name or the X, Y and Z dimensions of the second
tomogram
Name of model of points from the second tomogram
Finally, enter name of file for output of the transformation, or
Return for none
Inputs to the program when matching models alone are being used:
A blank line (Return) in place of the name of the first fiducial file
The limiting value for the maximum residual
Either the file name or the X, Y and Z size of the first tomogram
Name of model of points from the first tomogram
Either the file name or the X, Y and Z size of the second tomogram
Name of model of points from the second tomogram
Name of file for output of the transformation, or Return for none
HISTORY
Written by David Mastronarde, 1995; modified for zero shifts, 7/4/97
Added outlier elimination and error exit, 6/5/99
Added ability to start with small initial set of matches, 3/20/00
Added ability to use matching models only, 7/21/02
Fixed treatment of single-surface case and converted to PIP, 12/24/03
Added ability to deal with absolute fiducial coordinates, 6/9/04