scExplorer — Downstream

Optional interpretation modules

Downstream Analyses

Run heavier biological interpretation modules only after embedding and clustering are stable. Each module writes artifacts back to this UUID.

Loading analysis context...

Checking downstream workspace...

Modules 5

Completed 0

Running 0

Needs Attention 0

UUID Waiting for analysis UUID

Species Not checked

Embedding Not checked

SCENIC resources Not checked

Velocity layers Validated on run

Review exported artifacts

Module guide Match the biological question before running a heavy job

Rows stay synchronized with module status. Use them to choose the method, then jump to parameters.

Trajectory (PAGA + DPT)

Not run

Use for lineage-like progression where a meaningful root can be defined.

Requires completed clustering and neighborhood graph.
For biologically grounded pseudotime, set root strategy to a known early cluster.

Root strategy

Root cluster (optional)

RNA Velocity (scVelo)

Not run

Use to model short-term transcriptional dynamics and directional transitions between states.

Requires adata.layers["spliced"] and adata.layers["unspliced"].
Default mode is stochastic (author default in scVelo).

Mode

Basis

SCENIC (pySCENIC)

Not run

Use to infer gene regulatory programs (TF-regulons) and quantify AUCell activity per cell.

Requires pySCENIC resources (env vars or files under /app/resources/pyscenic).
Default minimum target genes per regulon is 20.

TF list

Min regulon genes

SCENIC Advanced Analysis

Blocked

Runs post-SCENIC analyses from existing regulon/AUCell outputs (no full pySCENIC rerun).

Waiting for SCENIC completion to load available groups/regulons.

Group by

Top regulons

Min cells/group

Binarization

Quantile

Z-score threshold

Run t-SNE

Optional and slower on large AUCell matrices.

t-SNE perplexity

t-SNE iterations

Network max edges

Network min weight

CellChat (Python / LIANA)

Not run

Use to estimate inter-cell communication edges from LR interaction scoring across clusters.

At least two groups with >= min_cells_per_group.
Database default auto resolves by species (human/mouse).

Database

Group by

Min cells/group

Downstream Help

Before You Start

Downstream analyses assume you already completed Upload, Preprocess, and Embedding. Use Check Readiness to validate UUID/session and embedding availability.

Recommended Order

1) Trajectory for state ordering, 2) Velocity for directionality, 3) SCENIC for regulatory programs, 4) CellChat for communication networks.

Trajectory (PAGA + DPT)

Produces pseudotime estimates. The most critical parameter is the root definition. If biological prior exists (early progenitor cluster), use root_strategy=cluster.

Output: pseudotime CSV + UMAP pseudotime plot.
Risk: wrong root leads to biologically inverted trajectories.

RNA Velocity (scVelo)

Requires spliced/unspliced layers. Default mode stochastic is robust for many datasets. Use dynamical only when data quality and depth justify model complexity.

Output: velocity stream plot (PNG, native scVelo) + interactive stream-like overlay (Plotly HTML) + interactive velocity score plot (HTML) + per-cell velocity score CSV.
Failure mode: missing layers or low kinetic signal.

SCENIC (pySCENIC)

Executes full CLI pipeline: grn -> ctx -> aucell. Requires configured resources in backend env vars (PYSCENIC_*).

Output: adjacencies, regulons, AUCell matrix, regulon activity UMAP.
Default: minimum regulon gene count = 20.

SCENIC Advanced Analysis

This post-processing block appears only after SCENIC completes. It reuses existing SCENIC outputs and adds regulon-level summaries, binarization QC, RSS-based specificity views, optional t-SNE in regulon space, and TF-target network views.

Output: AUCell/group heatmaps, binary activity heatmap, RSS heatmap + top RSS bars, optional t-SNE, and regulon network.
Cost: much lighter than full SCENIC; no GRN/CTX rerun.

CellChat (Python via LIANA)

Runs LIANA's CellChat method to score ligand-receptor communication between groups. Group sizes strongly affect stability; keep min_cells_per_group realistic.

Output: interaction table, source-target edges, CellChat probability heatmap, CellChat circle network, CellChat bubble plot, LIANA dotplot, and LIANA sender/receiver heatmap.
Default: min_cells_per_group=10, database auto by species.

Common Errors and Fixes

No UUID in session: run upload first in this browser session.
No artifacts found: module has not run yet or failed in SLURM.
SCENIC resource error: configure PYSCENIC_* variables or mount files in /app/resources/pyscenic; then re-run Check Readiness.
Velocity layer error: provide h5ad with spliced/unspliced layers.