.na
.nh
.TH pickbestseed 1 4.6.8 IMOD
.SH NAME
pickbestseed \- Selects best seed points for autofidseed
.SH SYNOPSIS
pickbestseed options
.SH DESCRIPTION
Pickbestseed is used by Autofidseed(1) to select an optimal set of seed points
fof the desired number, using bead models tracked starting from several
different views and using information on the quality of the tracking, the
shape of the beads, and on which of two surfaces they are located.
.P
The program operates with the following steps:
1) It reads in the different tracked models and identifies cases where the
same bead was tracked in different models, as signified by tracks that are
sufficiently close to each other over the whole range of views tracked
(controlled by parameters 1 and 6 below).
2) Each unique bead becomes a candidate, and a single score is obtained
by taking a weighted mean of four measures of the consistency and
quality of tracking: the mean residual from alignment during tracking;
the fraction of models in which the bead is missing; the fraction of
points missing from each track of the bead; and the mean distances
between corresponding points in the tracks, computed from each pair of
tracks. By default, these measures are give equal weighting. A lower
score is better. When beads are located on two surfaces, each
candidate is assigned to a surface based on which assignment
predominates in the multiple surface analyses available.
3) The total area to be analyzed is measured, so that it is possible to
convert between number of beads and average density per unit area.
4) Information on the elongation of beads is analyzed, namely statistics from
the standard deviation of pixels around the bead and for the elongation. An
adjusted value for the mean of the elongation over the 11 views of a track is
computed by, in effect, rotating a plot of mean elongation versus the standard
deviation of the elongation by an angle (parameter 18 below) that removes much
of the variability due to the SD of the elongation, and using the new Y
coordinate as the adjusted value. The same operation is performed with the SD
of pixels around the beads to obtain an adjusted edge SD (rotation angle
controlled by parameter 19). The latter values are scaled to have the same
standard deviation as the adjusted elongation values, and then a plot of
adjusted elongation versus scaled, adjusted SD is rotated so as to combine
them into a single measure (parameter 20 below), with a final elongation
measure taken as the Y coordinate after rotation. This measure is analyzed by
taking the median, finding the median absolute deviation, and considering
values as outliers if their deviation from the median is more than the
normalized median absolute deviation (MADN) times a criterion (parameter 12
below). A candidate is marked as elongated if it is an outlier or if the
median elongation exceeds an absolute threshold (parameter 16 below).
5) Beads are identified as clustered if they are within a criterion
distance (parameter 5 below) of any other bead on the same surface. This
distance is evaluated at the highest tilt angle of the series, assuming that
the distance between beads perpendicualr to the tilt axis is foreshortened by tilting.
6) Elongation is analyzed again, considering only points that are not
identified as clustered. When there are many clustered points, which also
tend to have high elongation values, this can skew the criterion for
identifying outliers, so analyzing unclustered points separately can identify
additional outliers. Each unclustered point is given the maximum of the
elongation score from the two analyses.
7) Beads are sorted in order by their overall score and two different
procedures are used to accept points in the final model, in a set of phases.
Only unclustered, unelongated beads are considered for the first 4
phases. The median and MADN of all the scores are used to define a
maximum acceptable score in this phase (8 MADNs from the median, by default).
Once some points have been accepted, the program computes a continuous 2D
density function from the points using kernel density estimation with a
triweight kernel. Specifically, for each bead, a component proportional to
(1 - (dist / H)^2)^3 is added for a point at distance "dist" from the bead. H
is chosen based on the target spacing to be achieved between points in the
current phase, times parameter 10 below.
7.1) Phase 1: The target density is converted to an equivalent spacing and beads are
accepted in order by their score, provided that they are not closer to
an already-added bead than a certain
fraction of this spacing (parameter 14 below). This procedure is then
repeated with the best half of the candidates, adding them if they are not
closer to an added bead than a lower fraction of the target spacing
(parameter 15 below). When there are beads on two surfaces, this
procedure is done separately for each surface.
7.2) Phase 2: A gap-filling routine is used to add further points up to a
desired density, if necessary. This routine repeatedly finds the point with
lowest density then searches out from that point in successively wider rings
for a bead to add. The ring spacing is the target spacing times parameter 3
below. If multiple beads are found in a ring, they are prioritized by an
adjusted score, which is their overall score divided by the distance to the
nearest accepted bead. In addition, if clustered and overlapped points are
being accepted (in a later phase), the score is increased for a clustered or
overlapped point, and a point within the clustering distance of another
accepted point is simply excluded. After a search is done in one location,
points within a certain distance of the density minimum are excluded from
further consideration on that call of the gap-filling routine. (This
criterion distance is the target spacing times parameter 2 below.) The search
is terminated when all density minima below a fraction of the target density
are examined. This fraction is parameter 8 below, but if there are two
surfaces and the ratio of minority to majority surface is less than parameter
13 below, it uses the higher fraction in parameter 9 instead for the minority
surface. The gap-filling routine is called twice, once allowing two rings,
then allowing the number of rings in parameter 4 below.
7.3) Phase 3: If there are two surfaces, it now tries to beef up the number on
the majority surface to make up for the deficiency. The target number for
this surface is the full target number minus the number on the minority
surface, unless the "-nobeef" option is entered, in which case the target is
still half the full target. First it calls the routine that considers points
in order by score and adds them if their distance from other points is high
enough. Then calls the gap-filling routine, but now density is computed from
points on both surfaces so that it can fill gaps left by the beads on the
minority surface preferentially. Again, the gap-filling routine is called
twice with two different numbers of rings.
7.4) Phase 4: If points are still deficient, it calls the gap-filling routine,
examining points with density up to the higher fraction (parameter 9) of the
target density. A revised target is used for the majority surface unless
"-nobeef" is entered, and the original target is used for the minority
surface. Densities are computed per surface, and the routine is called only
once with the full number of rings.
7.5) Phases 5-8: If clustered and/or elongated points are allowed to be included,
then it runs the same procedure as in phase 4, first allowing clustered points
if they are allowed, and then elongated points with progressively higher
elongation numbers, which are based on the fraction of tracked models in which
the bead was identified as elongated. In these phases, a larger
maximum score is allowed (12 MADNS above the median by default), since
most beads with high scores fall in these categories.
8) Accepted points are put into the output model, along with a general value
equal to the inverse of the score. With two surfaces, points on the top surface
are given surface number 1, which is assigned magenta color.
.SH OPTIONS
Pickbestseed uses the PIP package for input (see the manual page
for pip(1)). Options can be specified either as command line
arguments (with the -) or one per line in a command file
(without the -). Options can be abbreviated to unique letters; the currently
valid abbreviations for short names are shown in parentheses.
INSERT OPTION TEXT HERE
.TP
.B -StandardInput
Read parameter entries from standard input
.SH AUTHOR
.nf
David Mastronarde
.fi
.SH SEE ALSO
autofidseed(1)
.P
Email bug reports to mast@colorado.edu.