Process bulk samples

You can use the pipeline of your own choice to process bulk samples, the requirements from DeepDETAILS are:

1. BigWig files for the signals on the forward and reverse strands respectively.
2. Peak regions in bed format.

For example, you can use the pipeline here to process sequencing library from TSS-assays like PRO/GRO-cap.

If your pipeline doesn't produce bigWig files, but it generates BAM files, then your can use third-party tools, such as bamCoverage or bedtools to convert them into bigWig files. Example

bamCoverage -b input.bam -o pl.bw --binSize 1 --filterRNAstrand forward
bamCoverage -b input.bam -o mn.bw --binSize 1 --filterRNAstrand reverse

Process reference samples

You can use the pipeline of your own choice to process reference single-cell/single-nucleus ATAC-seq samples. After clustering cells and annotating the cell types of each cluster, you can follow examples below to export the fragments and annotation files required by DeepDETAILS.

# assume your Signac/Seurat object is `scexp`
# use cluster labels from non-linear dimension reduction and clustering
write.table(
  as.data.frame(scexp$seurat_clusters),
  'ann.tsv',
  sep = '\t',
  quote = F,
  col.names = F,
  row.names = T
)

# or if you have integrated labels from other scRNA-seq experiments
write.table(
  as.data.frame(scexp$predicted.id),
  'ann.tsv',
  sep = '\t',
  quote = F,
  col.names = F,
  row.names = T
)

# assume your ArchR project object is `proj`
meta_df <- getCellColData(proj, select='Clusters')

# Note: the cell barcode in ArchR maybe concatenated with the sample name
# In this case, the row names of meta_df looks like
# DeepDETAILS requires a strict match in the fragments file and the annotation
# so you may need to run the followling line to remove sample name
# rownames(meta_df) <- sub("^[^#]+#", "", rownames(meta_df))

write.table(
  meta_df, 'ann.tsv',
  sep = '\t',
  quote = F,
  col.names = F,
  row.names = T)

tail -n +2 Graph-Based.csv | sed 's/,/\t/g' > ann.tsv

If you want DeepDETAILS to perform a preflight check (removing cell types in the reference but not in the bulk), you need to tell DeepDETAILS where are the accessible regions. Depend on your pipeline, you can use

• peaks.bed, for 10x Cell Ranger ATAC
• atac_peaks.bed, for 10x Cell Ranger Arc
• other peak calls from conventional peak callers like MACS2.

Build DeepDETAILS datasets

DeepDETAILS stores datasets in hdf5 files. After you have the required input files ready, you can build a deepDETAILS dataset with the following command:

deepdetails prep-data \
    --bulk-pl bulk.pl.bw \
    --bulk-mn bulk.mn.bw \
    --regions bulk.peaks.bed \
    --fragments fragments.tsv.gz \
    --barcodes barcodes.tsv \
    --accessible-regions atac_peaks.bed \
    --save-to ./dataset \
    --genome-fa /refs/sequences/human/GRCh38_no_alt_analysis_set_GCA_000001405.15.fasta \
    --chrom-size /refs/refdata-cellranger-arc-GRCh38-2020-A-2.0.0/star/chrNameLength.txt

A successful execution of the above command will create the following files:

• data.h5: The data file for deconvolution
• regions.csv: Extended bulk peak regions and background regions
• scatac.norm.csv: Normalization factors for each cell type
• *.fragments.bw: Pseudo-bulk ATAC-seq tracks for each cell type
• background_sampling.png: Diagnostic plot for the background sampling process
• signatures.csv/png: When preflight is enabled, these files give you the signal distribution across the signature regions