Description General about weights Signals and signal weights Probes and probe weights Probesets and probeset weights Data points and data-point weights Arrays and array weights Channels and channel weights Combining signal weights into spot weights Combining signal weights into data-point weights Combining spot weights into data-point weights Restrictions Author(s)
For several normalization and calibration methods the estimation of the normalization (or calibration) function can be done with weights. Commonly, weights are proportional to a quality measure, that is, the less quality we assign to a signal, the less influence (weight) it should have on the estimation of calibration and normalization functions.
The definition of a weight is a single value in [0,1].
Weights outside this range and NA
s (missing values) are not allowed.
Below, we will define different entities such as signals, probes/spots, probesets, channels, arrays, and data-points. To any of these entities weights may be assigned.
A signal is a single value. A signal weight is a weight assigned to a signal. Thus, it is for entities within an array and never across/between arrays.
Example: In two-color microarray data, there are two signals
for each spot, i.e. the red or the green signals, and each of them
can be assigned a different signal weight.
Typically, such signal weights are represented by an Nx2 matrix
,
where N is the number of probes/spots on the array.
Example: In Affymetrix microarray data, which is single-channel
data, there is one signal per probe (in turn part of a probe set).
Each such probe can be assigned a signal weight.
Typically, such signal weights are represented by an Nx1 matrix
,
where N is the number of probes/spots on the array.
A probe is the smallest entity (not considering image pixels) on the array that measures the amount of hybridized samples in one or several channels.
Example: For two-color microarrays, a probe is a spot. Example: For Affymetrix arrays, a probe can be either a perfect match probe (PM) or a mismatch probe (MM).
A probe weight is a weight assigned to a probe/spot (not a probe set).
Example: For two-color data, the signals in the two channels for a given spot share the same probe weight.
Example: For four-color data, the signals in the four channels for a given spot share the same probe weight.
Example: For single-channel data such as Affymetrix data, the probe weight is identical to a signal weight.
Typically, above signal weights are represented by an Nx1 matrix
,
where N is the number of probes/spots on the array.
The probe weight of probe $i$ must be equal to the mean of its signal weights.
A probeset consists of a set of probes.
Example: For two-color microarrays, probesets are not defined. Example: For Affymetrix arrays, a probeset is the set of perfect match (PM) and mismatch (MM) probes corresponding to the same gene.
A probeset weight is a weight assigned to a probeset.
Since Affymetrix is single-channel arrays, typically the above probeset
weights are represented by an Nx1 matrix
, where N is the number of
probesets.
The probeset weight for probeset $j$ must be equal to the mean of its probe weights. (==signal weights) by averaging the probe weights for each probeset.
The definition of a data point depends on the context. It may be assigned to entities within an array, but also across/between arrays.
Example: (Paired data-point weight). In paired-channel normalization, such as curve-fit normalization (a.k.a. lowess intensity normalization), two and only two channels are normalized together at the same time, e.g. red and the green channels in two-color data, or two two single-channel data set obtained from two different Affymetrix arrays. Here a data point is constituted by two signals, e.g. X_i = (G_i,R_i). A data-point weight is assigned to the pair of signals corresponding to the same spot or gene, e.g. for lowess normalization a data-point weight is assigned to a log-ratio and a log-intensity.
Example: (Multi-channel data point weight). In multi-channel normalization, such as affine normalization or quantile normalization (within a singel array and/or across multiple arrays), each data point is consituted by multiple signals, e.g. for K two-color arrays it is X_i = (R[i,1],G[i,1],...,R[i,K],G[i,K]). To this data point, a data-point weight can be assigned.
Typically, above signal weights are represented by an Nx1 matrix
,
where N is the number of data points.
Data-points weights can be generated from signal or probe weights, by averaging them for each data point.
If not stated elsewise, arguments named weights
are assumed
to take data-point weights.
An array weight is a weight assigned to an array, that is, to the complete set of signals in all channels constituting an array.
By definition, a channel weight can never apply across/between arrays.
Constraints: The array weight should be equal to the average of the channel (and signal/probe/probeset) weights. Hence, for single-channel arrays, the array weight should be identical to the channel weight.
A channel weight is a weight assigned to a channel, that is, to the set of signals constituting a channel.
By definition, a channel weight can never apply across/between arrays.
Example: In two-color data, two channel weights can exist.
Example: In Affymetrix (single-channel) data, only one channel weights can exists and is therefore identical to an array weight.
Constraints: The channel weight should be equal to the average of all signal/probe/probeset weights in the channel.
Spot weights can be generated from signal weights.
For a given spot, the spot weight is calculated as the arithmetical mean of the signal weights.
Data-point weights can be generated from signal weights.
For a given data point, the data point weight is calculated as the arithmetical mean of the signal weights.
Data-point weights can be generated from spot weights.
For a given data point, the data point weight is calculated as the arithmetical mean of the spot weights.
Note, currently weights are only supported by the RGData
class.
The plan is to make this class the "main" class.
Currently, it is only methods that explicitly say they support weights which use weights. For all other methods, weights are ignored.
Henrik Bengtsson (http://www.braju.com/R/)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.