This method calculates the kernel matrices for pairs of cell types based on their distances and a range of sigma values. The formula of calculating kernel matrix is: $$K(x, y) = \exp\left(-\frac{\|x-y\|^2}{2 \sigma^2}\right)$$ The matrices are adjusted by clipping the upper quantile of the values to reduce the effect of outliers. The results are stored within the object.
Usage
computeKernelMatrix(
object,
sigmaValues,
lowerLimit = 1e-07,
upperQuantile = 0.85,
normalizeKernel = FALSE,
minAveCellNeighor = 2,
rowNormalizeKernel = FALSE,
colNormalizeKernel = FALSE,
verbose = TRUE
)
# S4 method for class 'CoProSingle'
computeKernelMatrix(
object,
sigmaValues,
lowerLimit = 1e-07,
upperQuantile = 0.85,
normalizeKernel = FALSE,
minAveCellNeighor = 2,
rowNormalizeKernel = FALSE,
colNormalizeKernel = FALSE,
verbose = TRUE
)
# S4 method for class 'CoProMulti'
computeKernelMatrix(
object,
sigmaValues,
lowerLimit = 1e-07,
upperQuantile = 0.85,
normalizeKernel = FALSE,
minAveCellNeighor = 2,
rowNormalizeKernel = FALSE,
colNormalizeKernel = FALSE,
verbose = TRUE
)Arguments
- object
A
CoProobject.- sigmaValues
A vector of sigma values used for kernel calculation.
- lowerLimit
The lower limit for the kernel function, default is 1e-7.
- upperQuantile
The quantile used for clipping the kernel values, default is 0.85.
- normalizeKernel
Whether to normalize the kernel matrix? Default = FALSE. Note that normalization will not affect any downstream analyses, it is for numerical stability and easier interpretation only.
What is the minimum average number of cell in the neighbor? This step is to help set up the expected sparsity of the kernel matrix. If a kernel sigma value is too small, this result in too few neighbors for most cells, resulting in an overly-sparse matrix that makes the parameter estimation hard. Thus, the sigma values that results in an overly-sparse matrix will be removed for later analysis.
- rowNormalizeKernel
Whether the kernel matrix will be row-wise normalized? Note that row or column wise normalization will result in an asymmetric result in skrCCA inference.
- colNormalizeKernel
Whether the kernel matrix will be column-wise normalized? Note that row or column wise normalization will result in an asymmetric result in skrCCA inference.
- verbose
Whether to output the progress and related information