SCENIC Gene Regulatory Network Analysis

Use this skill when the user wants to infer transcription factor (TF) regulatory networks, identify regulons (TF + target gene sets), score regulon activity per cell, or find master regulators for specific cell types. SCENIC reconstructs gene regulatory networks from scRNA-seq data using a 3-stage pipeline.

Overview: 3-Stage Pipeline

1. GRN inference — Predict TF → target gene links using RegDiffusion (deep learning, 10x faster than legacy GRNBoost2)
2. Regulon pruning — Validate links with cisTarget motif enrichment databases, keeping only direct targets
3. AUCell scoring — Quantify regulon activity per cell, enabling regulon-based clustering and cell type characterization

Prerequisites

Data requirements

• Raw counts (NOT log-transformed). RegDiffusion needs count-level variance structure.
• HVG-filtered to ~3000 genes for tractable runtime.
• Cell type annotations in adata.obs (for downstream RSS analysis).

Database downloads (CRITICAL — most common failure point)