New features:

• Added a batch correction algorithm to cellranger-atac aggr, intended for use when aggregating datasets that use different chemistries. See the aggr and algorithms pages for more details on how and when to use batch effect correction.

• Added UMAP as an alternate projection in Loupe and Cell Ranger ATAC output files.

• Added support for custom peaks with peaks on non-primary contigs.

• Minor cell calling improvements:

  • Added less stringent filtering of low-targeting barcodes.
  • Improved calling on multi-species samples.

Reference changes:

• Added support for overlapping gene annotations in cellranger-atac mkref.
• Cell Ranger ATAC 2.1 can only be run on references generated by cellranger-atac mkref version 2.1 or 2.0. Cell Ranger ATAC 2.0 can only be run on references that are constructed using cellranger-atac mkref version 2.0.

Web summary improvements:

• Enabled improved identification of the non-nucleosomal fragment fraction. The metric "fragments in nucleosome free regions" reported in the web_summary.html has been redefined as the fraction of high quality fragments smaller than 124 basepairs (previously, the threshold was 147 basepairs).
• Increased warning threshold for number of called cells from 10k to 15k.
• Added warning to web summary when forcing cell calls to unsupported levels.
• Added notification when user is using custom peaks or forcing cell calls.
• Added links to support pages.

Bug fixes:

• Added missing format version (VN:) to header of BAM file.
• Removed dependence on external Unix commands in BAM processing.
• The metric "Fraction of fragments overlapping any targeted region" reported in the web_summary.html was dropped in version 2.0 because some custom reference lacks DNase HS sites, or promoter/enhancer data, but the metric was still present in the aggr web_summary.html. This empty metric has been removed in v2.1.
• In v2.0, the on_target_fragments metric (from the aggr pipeline) was switched with the blacklist_region_fragments in the singlecell.csv file. This column switch was fixed in v2.1.

New wavelet-based peak caller:

• Eliminates large peaks much larger than 5kb. Peaks have a tighter size distribution around ~ 1kb.
• Improved detection of cluster-specific peaks.
• Improved reproducibility between technical replicates.
• Consistent performance across a range of cell loads.
• Fixes crashes in the signal-background fitting procedure.

Change in duplicate marking algorithm:

• Two read pairs are duplicates if they share the same start, end and cell barcode. Previously, only the start and end were used.
• Boosts median fragments per cell by as much as 25% at high cell loads.
• We no longer distinguish between PCR and sequencer duplicates.

Improved computational performance:

• Up to 4x faster, 0.5x disk requirements.
• Complete rewrite of read processing and differential accessibility analysis in Rust.
• Minimize disk I/O.

Change to pre-built references:

• The following additional annotation tracks have been removed:

  • blacklist.bed
  • dnase.bed
  • ctcf.bed
  • enhancer.bed
  • promoter.bed

• As a consequence, columns in the singlecell.csv output for the corresponding tracks are all zero. Additionally, the on_target_fragments column, which is a sum of TSS, DNAse, enhancer and promoter columns represents the TSS fragments. This value would be lower than expected for this reason.

Breaking changes to reference package structure:

• Cell Ranger ATAC 2.0 pipelines cannot be run with a version 1.2 reference. Cell Ranger ARC 1.0+ or Cell Ranger ATAC 2.0 references must be used.
• Restrictions on the number of contigs or primary contigs in the reference have been eliminated.
• Change to the config file format to construct a reference using mkref.
• Primary contigs are now defined to be the set of gene-containing contigs and cannot be specified by the user.
• Disabled support for URLs in the config file.
• Disabled support for GFF annotations, annotations must be in GTF format.
• Eliminated discrepancies between reference checks in mkref and preflight checks in count. Previously, it was possible to pass checks in mkref and fail checks in count.

Added header lines beginning with # to the fragments.tsv.gz and peaks.bed files that contain version, reference and sample information.

Eliminated the --downsample option and replaced by --subsample-rate.

Change to ATAC peak annotation:

• Annotate peaks using all genes provided in the reference GTF. Previously only certain gene types were used for annotation.
• The peak column is now split into three columns: chrom, start, end.
• When a peak has multiple gene annotations, the same peak appears in multiple rows with each annotation. Previously, each row represented one peak and multiple annotations were expressed using ; separators in the same row.

Change to CSV definition file format for cellranger-atac aggr: eliminated peaks column. Custom peaks may be specified using a --peaks argument at the command line.

Eliminated normalization mode signal from cellranger-atac aggr.

Loupe browser files now contain pre-computed K-means clustering for K=2-5 (previously K=2-10).

Loupe browser files generated by the pipeline can only be opened by Loupe browser version 5.0 or later.

The web_summary.html file output from cellranger-atac count has been updated to be functionally consistent with that from cellranger-arc count.

The PLSA algorithm is restricted to use one thread and computational performance is likely to be affected.

Change to metric names in summary.csv generated by cellranger-atac count.

Eliminated secondary alignments from the position-sorted BAM.