Skip to main content

WG2: Cell-type classification

4 Split data

For data integration and cell type classification purposes, it is recommended to split the data into smaller subsets that reflects a single homogeneous experimental unit. In the context of cell type classification we define an experimental unit as the group of cells sharing similar biological and technical variance components. A batch (e.g pool/well from a single cell experiment) represents the most common experimental unit.

Therefore, we recommend splitting the data for:

  1. Faster classification/integration. Splitting the data avoids extensive search for cell anchors and neighbors in the azimuth approach
  2. Parallelization. Each experimental unit can be processed and analyzed independently using multiple CPU workers
  3. Avoiding intra-batch effects in the query dataset. Cell differences driven by batch effects within the query dataset may influence the performance of the integration/alignment.

We recommend splitting the data by experimental units (e.g. batches). If the Seurat query object represents a single experimental unit, no splitting is necessary and this section can be skipped.

To split the data, we run the following code. First, let’s create an output directory

mkdir step1_split

Let’s assume we have a Seurat RDS object called query.RDS with a metadata column name called pool containing the values 1 and 2 corresponding to two independent pools. We can split the data by pool using the split.R module.

Let’s examine the parameter information:

singularity exec -B $PWD cell_classification.sif \ 
  Rscript /split.R --help
Usage: split.R [options]


Options:
        --file=CHARACTER
                RDS object file name

        --batch=CHARACTER
                Batch column to split the data

        --out=CHARACTER
                Output file name [default= split_object]

        --path=CHARACTER
                Output path to store results [default= .]

        -h, --help
                Show this help message and exit

We can now split the data as follows

singularity exec -B $PWD cell_classification.sif \
  Rscript /split.R \
  --file query.RDS \
  --batch pool \
  --out query \
  --path step1_split

split.R takes two main arguments. --file corresponds to the seurat object filename in .RDS format. --batch expects a string corresponding to a column name in the @metadata slot. A group of RDS files following the *{out}_{batch levels}.RDS* filename pattern are created. In this example, two files called query_1.RDS and query_2.RDS are stored in the step1_split directory.

We strongly recommend storing results in different directories to avoid file overwriting.