LISI

To assess whether clusters of cells in a single-cell RNA-seq dataset are well-mixed across some categorical variable (e.g. batch, technology, donor), we provide an algorithm for computing a Local Inverse Simpson’s Index (LISI).

Citation

Learn more about how we use LISI to measure single cell integration methods in the Harmony paper:

• Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods (2019)

Or see the freely available pre-print at bioRxiv.

Installation

Install the lisi R package with devtools:

install.packages("devtools")
devtools::install_github("immunogenomics/lisi")

Example

We can compute the LISI for each cell with these inputs:

• a matrix of cells (rows) and coordinates (PC scores, tSNE or UMAP dimensions, etc.)

• a data frame with categorical variables (one row for each cell)

Here is a small example that uuses the data provided with the lisi R package.

library(lisi)

head(X)
#>          X1           X2
#> 1 -5.043102  1.296301452
#> 2 -6.110305 -0.734483462
#> 3 -6.069576  0.009776544
#> 4 -4.552468  0.170646253
#> 5 -4.601112 -0.074033337
#> 6 -5.026126 -1.822653071

head(meta_data)
#>   label1 label2
#> 1      A      A
#> 2      A      A
#> 3      A      B
#> 4      A      A
#> 5      A      B
#> 6      A      B

table(meta_data$label1)
#> 
#>   A   B 
#> 200 200

table(meta_data$label2)
#> 
#>   A   B 
#> 201 199

res <- compute_lisi(X, meta_data, c('label1', 'label2'))
head(res)
#>     label1   label2
#> 1 1.925592 1.997943
#> 2 1.994034 1.988416
#> 3 1.959509 1.714524
#> 4 1.999995 1.801086
#> 5 1.960422 1.863254
#> 6 1.999498 1.984862

Each row in the output data frame corresponds to a cell from X. The score (e.g. 1.92) indicates the effective number of different categories represented in the local neighborhood of each cell. If the cells are well-mixed, then we might expect the LISI score to be near 2 for a categorical variable with 2 categories.

Learn more by running ?compute_lisi in R.