In immunogenomics/LISI: Local Inverse Simpson Index (LISI) for scRNAseq data

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

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)

head(meta_data)

table(meta_data$label1)

table(meta_data$label2)

res <- compute_lisi(X, meta_data, c('label1', 'label2'))
head(res)

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.

immunogenomics/LISI documentation built on July 30, 2020, 5:49 a.m.