Step 4: Demultiplexing and doublet detection (HTO track)
In Step 4 of the HTO track, Seurat’s implementation (MULTIseqDemux) of the tag assignment algorithm outlined in Multi-seq is used to demultiplex pooled samples and identify doublets according to the expression matrices of the sample-specific barcodes (McGinnis et al 2019).
The following parameters are adjustable for Step 4 (~/working_directory/job_info/parameters/step4_par.txt):
| Parameter | Default | Description |
|---|---|---|
| par_save_RNA | Yes | Whether or not to export an RNA expression matrix |
| par_save_metadata | Yes | Whether or not to export a metadata dataframe |
| par_seurat_object | NULL | If users already have a Seurat object(s), they may provide the path to a directory that contains an existing Seurat object(s) to initiate the pipeline at Step 4 |
| par_normalization.method | CLR | Method for normalizing the HTO assay |
| par_scale.factor | 1000 | Scale factor for scaling the HTO assay |
| par_selection.method | vst | Method for selecting the most variable features in the HTO assay |
| par_nfeatures | 5 | Number of features to select as top variable features for the HTO assay. This value is dependent on the number of sample specific barcodes used in the experiment |
| par_dims_umap | 5 | Number of dimensions to use as input features for uniform manifold approximation and projection (UMAP) of HTO assay |
| par_n.neighbor | 65 | Number of neighboring points to use in local approximations of manifold structure |
| par_dimensionality_reduction | Yes | Whether or not to perform linear dimensionality reduction on the HTO assay |
| par_npcs_pca | 30 | Total Number of principal components to compute and store for principal component analysis (PCA) of HTO assay |
| par_dropDN | Yes | Whether or not to remove predicted doublets and negatives from downstream analyses |
| par_label_dropDN | Doublet, Negative | Labels used to identify doublet and negative droplets |
| par_quantile | 0.9 | The quantile to use for droplet classification using MULTIseqDemux |
| par_autoThresh | TRUE | Whether or not to perform automated threshold finding to define the best quantile for droplet classification using MULTIseqDemux |
| par_maxiter | 5 | Maximum number of iterations to use if autoThresh = TRUE |
| par_RidgePlot_ncol | 3 | Number of columns used to display RidgePlots, which visualizes the enrichment of barcode labels across samples |
| par_old_antibody_label | NULL | If you wish to rename the barcode labels, first list the existing barcode labels in this parameter. old antibody labels can be identified in the "_old_antibody_label_MULTIseqDemuxHTOcounts" file produced by running Step 4 msd |
| par_new_antibody_label | NULL | If you wish to rename the barcode labels, list the new labels corresponding to the old labels listed in the parameter above |
To demultiplex the samples and identify doublets, the first step is to obtain the barcode labels used in the analysis by running the following command:
bash $SCRNABOX_HOME/launch_scrnabox.sh \
-d ${SCRNABOX_PWD} \
--steps 4 \
--msd T
Note: This step will produce the old_antibody_label_MULTIseqDemuxHTOcounts.csv file, which contains the names of the old HTO labels. The names of the HTO labels can be revised to be more descriptive in the execution parameters of this step (par_old_antibody_label; par_new_antibody_label)
Next, demultiplex the samples and identify doublets by running the following command:
bash $SCRNABOX_HOME/launch_scrnabox.sh \
-d ${SCRNABOX_PWD} \
--steps 4
The resulting output files are deposited into ~/working_directory/step4. For a description of the outputs see here.