createRef: Create reference hierarchy for computing weighted normalized...

In haowulab/Wind: Wind: weighted indexes for clustering evaluation

Description

This function takes an expression matrix and true cell classes, and returns two square (JxJ) matrices, where J is the number of classes provided by true cell class.

Usage

createRef(Y, classes)

Arguments

Y	The expression matrix. Rows are genes, columns are cells.
classes	True cell classes.

Details

The weights used in wNMI are computed based on the heights of the branches of the hierarchical tree of different cell types. We first compute mean expression profiles for all cell types, and then select 1000 marker genes using the same procedure as described in 'createWeights' function. A hierarchical tree is then constructed based on the mean expression from marker genes. We then obtain the tree height (from the bottom) at each branching point. The ratios of these heights to the maximum tree height are used as the weight in computing wNMI.

Value

A list with following fields:

Ref	A matrix for the reference cell type clustering when cutting the full hierarchical tree at different branching point. Each column represents a cut. For example, if a column is [1,1,1,2,2,3], it means the first 3 cell types are deemed in the same cluster, the next two are in another cluster, and the last cell type is in a cluster by itself.
weights	A vector for the relative tree heights at different cut. This will be used as weights in computing wNMI.
hc	An object of class 'hclust'. This is the hierarchical tree of the reference cell types.

Author(s)

Zhijin Wu <[e-mail]>, Hao Wu <[e-mail]>

Examples

data(Zhengmix8eq)
ctStruct = createRef(Y, trueclass)
plot(ctStruct$hc)

haowulab/Wind documentation built on Nov. 4, 2019, 1:27 p.m.