CellRanger网页报告浅蓝色部分

下面的内容是从10X cellranger 官网里面看到的关于cellranger3.0以后的版本网页报告中的浅蓝色部分如何计算得到的讲解。

Calling Cell Barcodes

Cell Ranger 3.0 introduces an improved cell-calling algorithm that is better able to identify populations of low RNA content cells, especially when low RNA content cells are mixed into a population of high RNA content cells. For example, tumor samples often contain large tumor cells mixed with smaller tumor infiltrating lymphocytes (TIL) and researchers may be particularly interested in the TIL population. The new algorithm is based on the EmptyDrops method (Lun et al., 2018).

The algorithm has two key steps:

1. It uses a cutoff based on total UMI counts of each barcode to identify cells. This step identifies the primary mode of high RNA content cells.（有效细胞鉴定）
2. Then the algorithm uses the RNA profile of each remaining barcode to determine if it is an “empty" or a cell containing partition. This second step captures low RNA content cells whose total UMI counts may be similar to empty GEMs.（空细胞[背景]还是细胞分区鉴定）

In the first step, the original Cell Ranger cell calling algorithm is used to identify the primary mode of high RNA content cells, using a cutoff based on the total UMI count for each barcode. Cell Ranger takes as input the expected number of recovered cells, N (see --expect-cells). Let m be the 99th percentile of the top N barcodes by total UMI counts. All barcodes whose total UMI counts exceed m/10 are called as cells in the first pass.（有效细胞得到的方法）
In the second step, a set of barcodes with low UMI counts that likely represent ‘empty’ GEM partitions is selected. A model of the RNA profile of selected barcodes is created. This model, called the background model, is a multinomial distribution over genes. It uses Simple Good-Turing smoothing to provide a non-zero model estimate for genes that were not observed in the representative empty GEM set.
Finally, the RNA profile of each barcode not called as a cell in the first step is compared to the background model. Barcodes whose RNA profile strongly disagrees with the background model are added to the set of positive cell calls. This second step identifies cells that are clearly distinguishable from the profile of empty GEMs, even though they may have much lower RNA content than the largest cells in the experiment.

参考链接：https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/algorithms/overview#cell_calling