-
The q-score is a quality measure for an array based on a subset (see
later) of the spots present on the array. It is calculated for a
spot-quality filter and can be used to guide selection of a cut-off
value for that filter to remove low quality spots.
-
It is very common that the expression of a gene is measured multiple
times, using different sequence probes on the same array. For example,
several different cDNA clones that map to the same Unigene cluster can
be present on one array. These clones typically are distributed
randomly across the array. Although two such clones may not give the
exact same intensity measurement for the same sample, the log-ratios
can be treated as parallel measurements for clones that are 'well
behaved'. The average spread of the log-ratios of clones within
clusterids may be used to assess the quality of an array. The greater
the spread of the log-ratios indicates a lower quality array while a
better quality array should have a tighter spread of values. The
q-score is a measure of the 'within clusterid' spread of the data and
can be calculated using the formula on Figure 1.
|
|
Figure 1: Q-score formula. nk: number
of clones in a clusterid; m: number of clusterids with
more than 1 clones; x: log-ratio measurement for feature
i of clusterid k minus the mean log-ratio
for clusterid k.
|
A lower value of this score means a narrower spread of the data and a
better quality array, while a higher value a wider spread of the data
and lower quality array.
In addition to its potential use for array quality assessment in
array-to-array comparisons, the score can be used for eliminating
lower quality spots for a single array. It can be expected that by
increasing the stringency of a suitable filter spots with lower
quality will be eliminated resulting in a better q-score for the
remaining clones that were not filtered out.
-
To calculate q-scores the program will need the following inputs:
- result set list
- genelist
- filter selections to eliminate lower quality spots
- ranges for each filter
- log-ratio values to calculate q-score
- whether to run program in background
Result sets: The arrays for which q-scores are calculated have
to be in a result set list in your loader/arraylists/
directory. Currently, only Genepix (or Scanalyze/SpotReader) arrays
can be used.
Genelists: You can use your own list of genes (cloneids) to
calculate the q-score. The genelist is expected to be in your
loader/genelists/ directory. In addition, there are common genelists
available for all users. These common genelists are displayed together
with the content of your genelists directory and have a 'COMMONLISTS:'
tag pre-pending their names. Among these common genelists are
geneslists with the extention '_all'. These genelists list all
cloneids in the database that are mapped to clusterids with more than
one cloneid annotated to them . The second type of common genelists
have the extention '_corr_MMDDYY' and are generated as a result of
further processing of the first set of common genelists. These latter
files contain a subset of the complete genelists: the cloneids that
don't correlate well to the other cloneids belonging to the same
clusterid, across all published data in the database, are
removed. Currently, only mouse and human arrays can be analyzed,
because clusterids are available for only these two organisms in the
database.
Filters: Select the filter you would like to use to eliminate
lower quality spots from the array. The program will calculate
q-scores for the range between 'Filter min value' to 'Filter max
value' by dividing the range into 500 equal steps by default. At each
subsequent filter value the spots whose value for the given filter is
smaller than the current filter value are eliminated and the q-score
is calculated for the remaining clones using the selected log-ratio
data.
A comparison of different filters on a small dataset gave the best
results using the correlation and 'error' filters. This analysis was
based on several criteria of the q-score vs fractions graph: the area
under the (the smaller the better); the initial slope of the curve
(the steeper the better); the smoothness of the curve (the smoother
the better).
Ranges: minimal and maximal values for the selected
filters. The accepted values are either numerical values or
'min'/'max'. In the latter case the program will find the minimal and
maximal values for the filter for the arrays in the arraylist and use
that as the range. 'Filter min value' is expected to be the least
stringent value for the filter; 'Filter max value', the most. For
filters that become more stringent with increasing numerical value
(e.g. regression correlation) 'Filter min value' can be a smaller
number or 'min' and 'Filter max value' a larger number or 'max'. For
filters that are more stringent at lower numerical values
(e.g. regression ratio) 'Filter min value' can be a larger number or
'max' and 'Filter max value' a smaller number or 'min'.
Log-ratio values: The values from this column are used to
calculate q-score.
Running program in background: Whether you would like the
program to run in the background. It is recommended that you use this
feature. When the calculation is over you will receive an email with a
link to your results.
-
After starting q-score calculation, the program first retrieves all
data from the database, putting all data in one pcl file for each
filter and log-ratio column selected. During this step very minimal
data filtering is done: flagged, contaminated, failed, empty and
control spots are removed. After data retrieval q-scores are
calculated for each filter and array at each filter value.
-
The program will produce the following five files for each filter
selected:
- result file (.res)
- Q vs Filter graph
- Fraction vs Filter graph
- Q vs Fraction graph
- new resultset list file
The files can be either directly navigated to in the browser or a link
is emailed to you if the process was run in the background.
Result file: This is a tab-delimited text file (.res):
- First column: filter values
- Second column: scores for array #1
- Third column: fractions for array #1
- Fourth column: scores for array #2...
The fractions are calculated relative to the initial number of
cloneids for which data could be retrieved from the database.
Q vs Filter: displays the q-scores as a function of filter
values (less to more stringent).
Fraction vs Filter: displays the fraction of clones not
filtered out as a function of filter values (less to more stringent)
Q vs Fraction: displays q-scores as a function of the fraction
of clones not filtered out.
-
The expectation is that the q-score will decrease more or less
monotonously with increasingly stringent filter values. Using a filter
that is removing lower quality spots you might expect the score to
drop faster initially, then reach a region of plateau or slower rate
of decrease. You might want to pick a filter value at this inflection
point. The graph with the remaining fraction of the clones gives you
an idea about what fraction of the clones are lost at the chosen
filter value. An example is shown on the Q vs Fraction graph
below (Figure 2.). The arrows show clear inflection points for
the arrays indicated. The corresponding filter values can be
determined from the Fraction vs Filter or the Q vs
Filter graphs.
-
- Currently, only GenePix (and ScanAlyze) arrays can be used.
- Only human and mouse arrays may be analyzed
- Number of arrays is not limited, but the graphs (meant only
to be a temporary solution) quickly become cluttered. If you are
happy with the text result file, you can analyze as many as you
like (known to have worked for ~180 arrays). If you need the
graphs, ~20 is the limit where you can still discern the
individual patterns.
