Regulatory Network Inference (SCENIC/REACTOR)

This page documents the SCENIC/REACTOR-based regulon analysis pipeline. See Methods for the manuscript prose.

Pipeline

Step 1 — Regulon Inference (GRNBoost2)

  • Input: raw-count scRNA-seq expression matrix
  • Runs: 40 independent GRNBoost2 runs
  • Pruning: motif-based using motifs-v9-nr.hgnc (m0.001)
  • CisTarget database: hg38 RefSeq r80 ±10 kb TSS (hg38__refseq-r80__10kb_up_and_down_tss.mc9nr)
  • Aggregation: regulons retained if recovered in ≥ 25% of runs

Step 2 — Earlier Draft Activity Scoring (AUCell) + Binarization => SCENIC essentially

  • Input: Single-cell counts
  • Purpose: reduce sparsity and improve robustness

Step 3 — REACTOR Donor-Cell-Type Activity Summaries

  • Source update: Hasan reported a new REACTOR analysis on 2026-06-22.
  • Input representation: binarized SCENIC regulons.
  • Aggregation level: donor-cell_type combinations.
  • Activity summary: for each donor-cell_type combination, REACTOR calculates the percentage of cells in which each regulon is active.
  • Interpretation: this percentage represents the fraction of cells in the donor’s cell-type population with active regulon signal

Step 4 — Differential Testing

  • Contrast: T2D vs Control.
  • Within-population comparison: percentages of active cells are compared between conditions within each population.
  • Test: ROTS.

Software

  • pySCENIC v0.12.1 todo check out
  • REACTOR
  • ROTS