"computeSumFactors"(x, sizes=c(20, 40, 60, 80, 100), clusters=NULL, ref.clust=NULL, positive=FALSE, errors=FALSE)
"computeSumFactors"(x, ..., assay="counts", get.spikes=FALSE)clusters to be used as the reference cluster for inter-cluster normalization.computeSumFactors,matrix-method.counts or exprs.computeSumFactors,matrix-method, a numeric vector of size factors for all cells in x is returned.For computeSumFactors,SCESet-method, an object of class x is returned containing the vector of size factors in sizeFactors(x).If errors=TRUE, the standard error of the size factor estimates is stored as the "standard.error" field of the attributes of the returned vector.
computeSumFactors function provides an implementation of the deconvolution strategy for normalization.
Briefly, a pool of cells is selected and the counts for those cells are summed together.
The count sums for this pool is normalized against an average reference pseudo-cell, constructed by averaging the counts across all cells.
This defines a size factor for the pool as the median ratio between the count sums and the average across all genes. Now, the bias for the pool is equal to the sum of the biases for the constituent cells.
The same applies for the size factors (which are effectively estimates of the bias for each cell).
This means that the size factor for the pool can be written as a linear equation of the size factors for the cells.
Repeating this process for multiple pools will yield a linear system that can be solved to obtain the size factors for the individual cells. In this manner, pool-based factors are deconvolved to yield the relevant cell-based factors.
The advantage is that the pool-based estimates are more accurate, as summation reduces the number of stochastic zeroes and the associated bias of the size factor estimate.
This accuracy will feed back into the deconvolution process, thus improving the accuracy of the cell-based size factors.
The standard error of the estimates can be obtained by setting errors=TRUE.clusters argument where cells in each cluster have similar expression profiles.
Deconvolution is subsequently applied on the cells within each cluster.
Each cluster should contain a sufficient number of cells for pooling -- twice the maximum value of sizes is recommended.
A convenince function quickCluster is provided for rapid clustering based on Spearman's rank correlation. Size factors computed within each cluster must be rescaled for comparison between clusters.
This is done by normalizing between clusters to identify the rescaling factor.
One cluster is chosen as a ``reference'' (by default, that with the median of the mean per-cell library sizes is used) to which all others are normalized.
Ideally, a cluster that is not extremely different from all other clusters should be used as the reference.
This can be specified using ref.clust if there is prior knowledge about which cluster is most suitable, e.g., from PCA or t-SNE plots.sizes to obtain an over-determined system that can be solved with methods like the QR decomposition. In theory, it is possible to obtain negative estimates for the size factors.
These are most likely for very small library sizes and are obviously nonsensical.
Some protection can be provided by setting positive=TRUE, which will use linear inverse models to solve the system.
This ensures that non-negative values for the size factors will always be obtained.
Note that some cells may still have size factors of zero and should be removed prior to downstream analysis.
Such occurrences are unavoidable -- rather, the aim is to prevent negative values from affecting the estimates for all other cells. By default, get.spikes=FALSE which means that spike-in transcripts are not included in the set of genes used for deconvolution.
This is because they can behave differently from the endogenous genes.
Users wanting to perform spike-in normalization should see computeSpikeFactors instead.quickCluster
set.seed(100)
popsize <- 800
ngenes <- 10000
all.facs <- 2^rnorm(popsize, sd=0.5)
counts <- matrix(rnbinom(ngenes*popsize, mu=all.facs*10, size=1), ncol=popsize, byrow=TRUE)
out.facs <- computeSumFactors(counts)
Run the code above in your browser using DataLab