Single-Cell RNAseq Analysis v5.0.0

Cellranger v2.0.0
Cellbender v1.0.1
Counts to adata v1.2.0
Cohort analysis v5.0.0

What's Changed

• Use the Siletti whole human brain taxonomy for human MapMyCells (new required mmc_taxonomy input: SEAAD, Siletti or ABC); DA neurons (cluster Splat_395) are labeled "Dopaminergic"
• Optional Kamath et al. 2022 DA neuron spike-in (kamath_post_qc_adata_object) so scANVI can assign DA subtypes; Kamath DA marker genes are always kept in the HVGs
• Optional Siletti et al. 2023 midbrain spike-in (siletti_post_qc_adata_object), spiked in unlabeled to benchmark scANVI label transfer against ground truth (${cohort_id}.spike_in_benchmark.csv)
• New obs["scanvi_labels"] column for scANVI training labels; obs["cell_type"] keeps the MapMyCells call
• Per-sample MAD-based QC thresholds for total counts and genes detected (n_mads_lower/n_mads_upper), computed only on barcodes within the absolute limits (UMI max 150,000, genes max 15,000)
• Save filtered-out cells (${cohort_id}.filtered_out_cells.csv)
• Add brain region levels 1–3 to the sample struct and AnnData objects; unset sex and brain region default to "NA"
• Make downsampling optional (n_cells); all cells are kept by default
• Add modality (rna / atac / multiome) to the raw and staging data paths so multimodal outputs don't overwrite each other (for sc-multiome-wf)
• Use rapids-singlecell (GPU) for neighbors, Leiden, UMAP and Harmony; exact kNN and spectral UMAP initialization
• scVI/scANVI: gradient clipping, input and NaN checks, higher epoch floors (scVI 100, scANVI 20) with early stopping
• Run scib metrics on a 500k-cell subsample, excluding spike-in cells
• Fix MapMyCells running out of memory; more threads, memory and disk for map_cell_types
• LZF compression for h5ad outputs; SSD disks for all tasks
• sc_tools:1.3.0: Python 3.12, pinned constraints.txt, rapids-singlecell 0.18.0; drop muon, harmonypy and scrublet
• Add a user guide to the README
• Feature/sc multiome by @kfang4 in #104
• Feature/sc revamp dec major release 2026 by @kfang4 in #105
• Feature/siletti spike in by @kfang4 in #106
• Feature/adjust compute resources full run by @kfang4 in #108

Full Changelog: sc_rnaseq_analysis-v4.1.0...sc_rnaseq_analysis-v5.0.0

Single-Cell RNAseq Analysis v4.1.0

Cellranger v2.0.0
Cellbender v1.0.1
Counts to adata v1.2.0
Cohort analysis v4.0.0

What's Changed

• Update Cell Ranger to v10.1.0
• Output BAM files and out dir from Cell Ranger
• Temporarily rename FASTQs (unique) for Cell Ranger if technical replicates are present for the sample (e.g., different runs for same sample, then increment the S)
• Allow scib-metrics to run even with one batch
• Add sex to adata.obs
• Specify cpuPlatform in WDL
• Name sample-level output files with dataset ID (because there are occurrences where the same biological sample is used in different experiments/datasets)
• Allow pmdbs_sc_multiome as a valid workflow name for sc-multiome-wf
• Feature/update preprocessing minor release sep 2026 by @kfang4 in #102

Full Changelog: sc_rnaseq_analysis-v4.0.0...sc_rnaseq_analysis-v4.1.0

Single-Cell RNAseq Analysis v4.0.0

Cellranger v1.1.0
Cellbender v1.0.0
Counts to adata v1.1.0
Cohort analysis v4.0.0

What's Changed

• This workflow now takes in PMDBS sc, PMDBS mulitmodal sc, and mouse sc data
• Change GitHub repo name from pmdbs-sc-rnaseq-wf to sc-rnaseq-wf
• Add workflow name mouse_sc_rnaseq in addition to pmdbs_sc_rnaseq
• Python scripts were updated to accommodate mouse data:
  • Use cell_type_mapper.cli.from_specified_markers for mouse
  • Score cell cycle for both human and mouse
  • Summarize MapMyCells results for both human and mouse
• Accommodate FASTQ file naming for R3 FASTQs
• Choose parameters when using scvi-tools if numerical instabilities predicted

Full Changelog: pmdbs_sc_rnaseq_analysis-v3.1.0...sc_rnaseq_analysis-v4.0.0

PMDBS Single-Cell RNAseq Analysis v3.1.0

Cellranger v1.1.0
Cellbender v1.0.0
Counts to adata v1.1.0
Cohort analysis v3.1.0

What's Changed

• Accommodate for processing single-cell multimodal data by adding --chemistry=ARC-v1 flag to cellranger count if multimodal_sc_data true
• Choose parameters when using scvi-tools if numerical instabilities predicted
• When labelling cell types with scANVI, set DataLoader num_workers = 0 due to Pytorch bug unable to mmap solution (pytorch/pytorch#92134)
• Cap memory selection for GCP N1 high-mem machines (default) and select GCP N2D high-mem machines with cpuPlatform: "AMD Rome" for memory-extensive task like the normalize task
• Use own structs.wdl instead of using wf-common's to add Boolean multimodal_sc_data in Project struct

Full Changelog: pmdbs_sc_rnaseq_analysis-v3.0.0...pmdbs_sc_rnaseq_analysis-v3.1.0

PMDBS Single-Cell RNAseq Analysis v3.0.0

Cellranger v1.1.0
Cellbender v1.0.0
Counts to adata v1.1.0
Cohort analysis v3.0.0

What's Changed

• Deprecate assigning cell types with marker list and implement Allen Institute's MapMyCells
• Add scANVI after scVI
• Revert back to running cross-team cohort analysis with scib for validation metrics after fixing program resource limits
• Rename scvi docker image to sc_tools
• Update Python packages compatible with Python 3.10.12
• Update wdl-ci to v2.1.0

Full Changelog: pmdbs_sc_rnaseq_analysis-v2.2.0...pmdbs_sc_rnaseq_analysis-v3.0.0

PMDBS Single-Cell RNAseq Analysis v2.2.0

Cellranger v1.1.0
Cellbender v1.0.0
Counts to adata v1.1.0
Cohort analysis v2.2.0

What's Changed

• Update manifest files to include md5 hash and GitHub release link, and group subworkflow versions as one main workflow version
• Update bucket paths to reflect bucket restructure
• Use wf-common for common WDL tasks/workflows, scripts, and docker images
• Update default groups to include ['sample', 'batch', 'cell_type', 'leiden_res_0.05', 'leiden_res_0.10', 'leiden_res_0.20', 'leiden_res_0.40']
• Update UMAP clustering to curate additional leiden clusters at multiple resolutions [0.05, 0.1, 0.2, 0.4] in order to enable post-hoc cell-typing
• Update CellAssign training parameters
• Save initial metdata
• Save additional feature metadata including all genes and highly variable gene (HVG) genes
• Only run team-level cohort analysis with scib for validation metrics due to program resource limits

Full Changelog: harmonized_pmdbs_analysis-v1.1.0_1.0.0_2.1.0...pmdbs_sc_rnaseq_analysis-v2.2.0

Harmonized PMDBS Analysis v.1.1.0_1.0.0_2.1.0

Cellranger v1.1.0
Cellbender + Counts to adata v1.0.0
Cohort analysis v2.1.0

• Converted workflow from R to Python
• Used Cellbender to remove technical replicates
• Implemented scib for validation metrics

Preprocess v1.1.0_1.0.0

Cellranger v1.1.0
Counts to seurat v1.1.0

Cohort analysis v2.0.0

• Update resources to enable processing of larger cohorts
• Reorganize output files

Preprocessing v1.0.0_1.0.0

Cellranger v1.0.0
Counts to seurat v1.0.0