Multi-modal profiling
The multimodal_profiling pipeline supports dependency-gated CITE-seq, ATAC, multiome, and cytometry workflows.
CITE-seq
Tabular ADT inputs are normalized with CLR and summarized with:
- ADT QC
- protein heatmap values
- protein-informed clusters
- joint RNA/protein coordinate fallback when paired
rna_/gene_andadt_/protein_features are supplied - optional per-cell isotype/background correction
- RNA/protein discordance summaries for immune marker programs
- CITE-seq phenotype panels using CD marker combinations
Full WNN graph construction and clustering remain dependency-gated. The current fallback is intended for inspectable research summaries, not a replacement for full multimodal graph learning.
ATAC
Tabular peak inputs produce:
- peak accessibility track rows
- total accessibility summaries
- TF motif enrichment cards
- per-cell accessibility QC with detected peaks, fragment/TSS/nucleosome proxies, and doublet-risk flags when columns are present
- dependency-free TF-IDF/LSI coordinates
- marker peaks by group
- gene activity scores from peak-to-gene annotations
- peak-to-gene correlation links
- group coverage rows
Full fragment-file processing, peak calling, motif deviation scoring, footprinting, and genome-browser tracks remain gated behind approved scATAC runtimes such as SnapATAC2.
Multiome
The current workflow provides an integration summary using protein clusters and accessible peaks. Full WNN/multiome graph integration remains gated behind muon.
Cytometry
CSV/TSV event tables support:
- automated lineage marker gating
- population frequencies
- cytometry coordinate fallback
- FlowSOM-style arcsinh-transformed prototype clustering
- metacluster assignments and minimum-spanning-tree edges
- marker heatmap rows by metacluster
- sample-level metacluster abundance summaries
Binary FCS parsing remains gated behind FlowUtils or flowutils.