Assume that datE.Admixture provides the expression values from a mixture of cell types (admixed
population) and you want to estimate the proportion of each pure cell type in the mixed samples (rows of
datE.Admixture). The function allows you to do this as long as you provide a data frame
MarkerMeansPure that reports the mean expression values of markers in each of the pure cell types.
proportionsInAdmixture(
MarkerMeansPure,
datE.Admixture,
calculateConditionNumber = FALSE,
coefToProportion = TRUE)is a data frame whose first column reports the name of the marker and the
remaining columns report the mean values of the markers in each of the pure populations. The function will
estimate the proportion of pure cells which correspond to columns 2 through of
dim(MarkerMeansPure)[[2]] of MarkerMeansPure. Rows that contain missing values (NA) will be
removed.
is a data frame of expression data, e.g. the columns of datE.Admixture could
correspond to thousands of genes. The rows of datE.Admixture correspond to the admixed samples for
which the function estimates the proportions of pure populations. Some of the markers specified in the
first column of MarkerMeansPure should correspond to column names of datE.Admixture.
logical. Default is FALSE. If set to TRUE then it uses the kappa
function to calculates the condition number of the matrix MarkerMeansPure[,-1]. This allows one to
determine whether the linear model for estimating the proportions is well specified. Type help(kappa)
to learn more. kappa() computes by default (an estimate of) the 2-norm condition number of a matrix
or of the R matrix of a QR decomposition, perhaps of a linear fit.
logical. By default, it is set to TRUE. When estimating the proportions the
function fits a multivariate linear model. Ideally, the coefficients of the linear model correspond to the
proportions in the admixed samples. But sometimes the coefficients take on negative values or do not sum to
1. If coefToProportion=TRUE then negative coefficients will be set to 0 and the remaining
coefficients will be scaled so that they sum to 1.
A list with the following components
data frame that contains the predicted proportions. The rows of PredictedProportions correspond to the admixed samples, i.e. the rows of datE.Admixture. The columns of PredictedProportions correspond to the pure populations, i.e. the columns of MarkerMeansPure[,-1].
data frame of numbers that is analogous to
PredictedProportions. In general, datCoef will only be different from PredictedProportions if coefToProportion=TRUE. See the description of coefToProportion
This is the condition number resulting from the kappa function. See the description of calculateConditionNumber.
vector of character strings that contains the subset of marker names (specified in the first column of MarkerMeansPure) that match column names of datE.Admixture and that contain non-missing pure mean values.
The methods implemented in this function were motivated by
the gene expression deconvolution approach described by Abbas et al (2009), Lu et al (2003), Wang et al (2006). This approach can be used to predict the proportions of (pure) cells in a complex tissue, e.g. the proportion of blood cell types in whole blood. To define the markers, you may need to have expression data from pure populations. Then you can define markers based on a significant t-test or ANOVA across the pure populations. Next use the pure population data to estimate corresponding mean expression values. Hopefully, the array platforms and normalization methods for datE.MarkersAdmixtureTranspose and MarkerMeansPure are comparable. When dealing with Affymetrix data: we have successfully used it on untransformed MAS5 data.
For statisticians: To estimate the proportions, we use the coefficients
of a linear model. Specifically:
datCoef= t(lm(datE.MarkersAdmixtureTranspose ~MarkerMeansPure[,-1])$coefficients[-1,])
where datCoef is a matrix whose rows correspond to the mixed samples (rows of datE.Admixture) and the columns correspond to pure populations (e.g. cell types), i.e. the columns of MarkerMeansPure[,-1].
More details can be found in Abbas et al (2009).
Abbas AR, Wolslegel K, Seshasayee D, Modrusan Z, Clark HF (2009) Deconvolution of Blood Microarray Data Identifies Cellular Activation Patterns in Systemic Lupus Erythematosus. PLoS ONE 4(7): e6098. doi:10.1371/journal.pone.0006098
Lu P, Nakorchevskiy A, Marcotte EM (2003) Expression deconvolution: a reinterpretation of DNA microarray data reveals dynamic changes in cell populations. Proc Natl Acad Sci U S A 100: 10370-10375.
Wang M, Master SR, Chodosh LA (2006) Computational expression deconvolution in a complex mammalian organ. BMC Bioinformatics 7: 328.