clusterClass: Clusters cells into different classes using PCA-Gaussian...
In kieranrcampbell/SpatialStats: Statistics for spatially resolved proteomics data

Description Usage Arguments Details Value References

A method of using high-dimensional proteomics data to classify cellular phenotypes (e.g. epithelial and stromal cells may be present within a tumour biopsy). First principle component analysis is performed on the raw cell-by-channel matrix Y. Then the top n.pc (default 3) principle components are used in Gaussian mixture modelling through expectation maximisation, which clusters the cellular readouts into nclass (default 2) different classes. The underlying assumption is that in the dimensionality-reduced PC-space the cells will be phenotypically distinct enough to appear drawn from two separate classes.

1	ClusterClass(Y, doPCA = TRUE, nclass = 2, n.pc = 3)

`Y`	A cell by channel matrix of readouts (e.g. from `cells(sp)`)
`doPCA`	If true, first performs PCA and uses the top principle components for clustering
`nclass`	The number of classes into which to cluster the cells (default = 2)
`n.pc`	The number of principle components to use in Gaussian mixture modelling (default = 3)

This is an unsupervised method - other methods may be used (e.g. training a SVM on known cell phenotypes). The cell classes can always be set by alternative methods using cellClass(sp) <- classes for an SPData sp and a vector of cell classes classes.

A vector of length number of cells with numeric values corresponding to distinct classes for each cell.

Chen, W.-C., Maitra, R., Melnykov, V. (2012) EMCluster: EM Algorithm for Model-Based Clustering of Finite Mixture Gaussian Distribution. R Package, URL http://cran.r-project.org/package=EMCluster

kieranrcampbell/SpatialStats documentation built on May 20, 2019, 9:24 a.m.