Fields and Methods
Methods:
rll{
getAsteriskTags -
}
Methods inherited from ProbeLevelModel:
calculateResidualSet, calculateWeights, fit, getAsteriskTags, getCalculateResidualsFunction, getChipEffectSet, getProbeAffinityFile, getResidualSet, getWeightsSet
Methods inherited from MultiArrayUnitModel:
getListOfPriors, setListOfPriors, validate
Methods inherited from UnitModel:
findUnitsTodo, getAsteriskTags, getFitSingleCellUnitFunction
Methods inherited from Model:
fit, getAlias, getAsteriskTags, getDataSet, getFullName, getName, getPath, getRootPath, getTags, setAlias, setTags
Methods inherited from Object:
asThis, getChecksum, $, $<-, [[, [[<-, as.character, attach, attachLocally, clearCache, clearLookupCache, clone, detach, equals, extend, finalize, gc, getEnvironment, getFieldModifier, getFieldModifiers, getFields, getInstantiationTime, getStaticInstance, hasField, hashCode, ll, load, objectSize, print, registerFinalizer, saveModel
For a single unit group, the log-additive model of RMA is:
$$log_2(y_{ik}) = \beta_i + \alpha_k + \varepsilon_{ik}$$
where $\beta_i$ are the chip effects for arrays $i=1,...,I$,
and $\alpha_k$ are the probe affinities for probes $k=1,...,K$.
The $\varepsilon_{ik}$ are zero-mean noise with equal variance.
The model is constrained such that $\sum_k{\alpha_k} = 0$.
Note that all PLM classes must return parameters on the intensity scale.
For this class that means that $\theta_i = 2^\beta_i$ and
$\phi_k = 2^\alpha_k$ are returned.Different flavors of model fitting
There are a few differ algorithms available for fitting the same
probe-level model. The default and recommended method
(flavor="affyPLM") uses the implementation in the
preprocessCore package which fits the model parameters robustly
using an M-estimator (the method used to be in affyPLM).
Alternatively, other model-fitting algorithms are available.
The algorithm (flavor="oligo") used by the oligo package,
which originates from the affy packages, fits the model using
median polish, which is a non-robust estimator. Note that this algorithm
does not constraint the probe-effect parameters to multiply to one on
the intensity scale. Since the internal function does not return these
estimates, we can neither rescale them.