In obricard/cleanhop: Index hopping correction for scRNAseq

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

Index hopping

Index hopping is a phenomenon in which index sequences are physically changed in the early steps of template amplification arising on the sequencing flow lane. This can be attributed either to residual index adapters that were not properly purified out or to degraded library molecules that contain index sequences. This material binds intact library molecules during their amplification on the flow lane and leads to a change of the index sequence, like what is obtained with a mutagenesis primer.
In the context of combinatorial dual index barcoding, index sequencing errors and index hopping are more likely to result in read sample misassignment if the donor sample/cell share already shares one index with the recipient sample/cell.

Cleanhop

Cleanhop aims to remove potential misassigned reads due to index hopping in single cell RNAseq data generated with combinatorial dual indexes (e.g.Nextera XT). For each cell and for each gene, it subtracts from the number of reads a percentage (0.5% by default) of the sum of the reads associated with the gene among all the cells sharing the same i7 index (column), and the sum of all the reads associated with the gene among all the cells sharing the same i5 index (row).
This treatment might mask the true expression of genes that are expressed at <100 lower level than other cells in the same row/column. However, index hopping anyway prevents any confident conclusion regarding such expression.

Usage

Cleanhop is intented to be use on SingleCellExperiment (sce) class object.
colData must contain columns specifing i5 and i7 identities.
if sce contains cells which have been sequenced on multiples flow lanes or in multiples run, colData must contain columns specifing flow lane identities and/or run identities.
Value of substracted percentage is set by default to 0.5% but can be adjusted for your dataset as index hopping likelyhood is dependant of library quality.