scRBP 0.1.1

Single-cell RNA Binding Protein Regulon Inference

scRBP — Single-cell RNA Binding Protein regulon inference

scRBP is a command-line toolkit for comprehensive analysis of RNA-binding proteins (RBPs) in single-cell RNA-seq data. scRBP provides a systematic, scalable and integrative framework to infer RBP-mediated gene and isoform regulatory networks (“regulons”) from single-cell transcriptomes and prioritize networks underlying complex genetic traits and disorders. scRBP is comprised of six main modules: (i) developing a comprehensive compendium of RBPs and their associated motif clusters from diverse public resources; (ii) systematic, motif-guided transcriptome-wide inference of RBP targets at both gene- and isoform-level resolution; (iii) construction of RBP-gene and/or RBP-isoform co-expression networks from short- or long-read single-cell transcriptomic data, respectively; (iv) defining high-fidelity regulons by integrating RBP-target interactions, and quantifying cell type-specific regulon activity scores (RAS); (v) integrating GWAS results to compute regulon-level genetic association scores (RGS); and (vi) constructing a unified trait-relevance score (TRS) by combining RAS and RGS for each regulon in a given cellular context, with statistical significance assessed using Monte Carlo (MC) sampling.

---

What scRBP Does

RBPs are key post-transcriptional regulators that control mRNA splicing, stability, and translation. scRBP enables you to:

• Construct which RBPs regulate which genes or isoforms in your single-cell data
• Prune raw RBP–gene associations using motif-binding evidence to obtain high-confidence regulons
• Score each cell or cell type for regulon activity score (RAS) using the AUCell algorithm
• Link RBP regulons to human disease through GWAS genetic enrichment (RGS via MAGMA)
• Integrate RAS and RGS into a unified Trait Relevance Score (TRS) that ranks disease-relevant RBPs

---

Pipeline at a Glance

Raw single-cell data (.h5ad / .feather)
          │
          ▼
[Step 1]  scRBP getSketch        ── Stratified GeoSketch cell downsampling
          │
          ▼
[Step 2]  scRBP getGRN           ── GRNBoost2/GENIE3 RBP→Gene/Isoform inference
          │                          (run N seeds for robustness, default 30 times)
          ▼
[Step 3]  scRBP getMerge_GRN     ── Merge N-seed GRNs → consensus network
          │
          ▼
[Step 4]  scRBP getModule        ── Extract regulon candidates (Top-N / percentile)
          │
          ▼
[Step 5]  scRBP getPrune         ── Motif-enrichment pruning via ctxcore
          │
          ▼
[Step 6]  scRBP getRegulon       ── Export pruned regulons to GMT format
          │
          ▼
[Step 7]  scRBP mergeRegulons    ── Merge region-specific GMT files
          │                          (3'UTR / 5'UTR / CDS / Introns)
          ▼
[Step 8]  scRBP ras              ── Regulon Activity Score (AUCell) per cell / cell type
          │
          ▼
[Step 9]  scRBP rgs              ── Regulon Gene-Set analysis (MAGMA GWAS enrichment)
          │
          ▼
[Step 10] scRBP trs              ── Trait Relevance Score (RAS × RGS integration)

---

Installation

Requirements

• Python 3.9, 3.10, or 3.11 (Python 3.12+ not yet supported by pyscenic/arboreto)
• MAGMA binary (external, required only for Step 9 — scRBP rgs)

Option 1 — Install from PyPI (recommended)

pip install scRBP

This installs scRBP together with all Python dependencies in one step.

Option 2 — Install from source (development)

git clone https://github.com/mayunlong89/scRBP.git
cd scRBP/scRBP_package
pip install -e .

Option 3 — Install via conda (recommended for HPC / cluster)

git clone https://github.com/mayunlong89/scRBP.git
cd scRBP/scRBP_package

conda env create -f environment.yml
conda activate scrbp

pip install -e .

Install MAGMA (for Step 9 only)

MAGMA is a standalone binary not available on PyPI. Download from https://cncr.nl/research/magma and make it executable:

# Linux example
wget https://cncr.nl/research/magma/software/magma_v1.10_static_linux.zip
unzip magma_v1.10_static_linux.zip -d ~/tools/magma/
chmod +x ~/tools/magma/magma

Verify installation

scRBP --help
scRBP getGRN --help

---

Quick Start

Step 1 — Downsample cells with GeoSketch

Large single-cell datasets (>500K cells) should be downsampled before GRN inference. scRBP uses GeoSketch to retain transcriptional diversity while reducing cell count.

scRBP getSketch \
    --input  PBMC_full.h5ad \
    --output PBMC_sketch_15K.feather \
    --n_cells 15000 \
    --celltype_col celltype \
    --min_cells_per_type 500 \
    --n_pca 100 \
    --seed 42

Step 2 — Infer gene regulatory networks (GRN)

Run GRNBoost2 with multiple random seeds. Each seed produces an independent GRN. Later, these are merged into a consensus network.

For this step, user can run 'getGRN' based on 30 random seeds for robustness.

Gene mode (RBP → Gene):

for SEED in $(seq 1 30); do
  scRBP getGRN \
      --matrix    PBMC_sketch_15K.feather \
      --rbp_list  human_RBP_list.txt \
      --output    grn_seed${SEED} \
      --mode      gene \
      --method    grnboost2 \
      --n_workers 20 \
      --correlation True \
      --seed      ${SEED}
done
# Output: grn_seed1_scRBP_gene_GRNs.tsv, grn_seed2_scRBP_gene_GRNs.tsv, ...

Isoform mode (RBP → Isoform, requires isoform annotation):

scRBP getGRN \
    --matrix                     PBMC_isoform.feather \
    --rbp_list                   human_RBP_list.txt \
    --output                     iso_grn_seed1 \
    --mode                       isoform \
    --isoform_annotation         gencode_v44_isoform_gene_map.tsv \
    --rbp_agg_method             sum \
    --remove_self_targets        True \
    --min_target_cells_expressed 10 \
    --min_target_mean_expr       0.01 \
    --method                     grnboost2 \
    --n_workers                  20 \
    --seed                       1
# Output: iso_grn_seed1_scRBP_isoform_GRNs.tsv (+ 4 auxiliary files)

Step 3 — Merge GRN seeds into a consensus network

scRBP getMerge_GRN \
    --pattern "grn_seed*_scRBP_gene_GRNs.tsv" \
    --output  grn_consensus.tsv \
    --n_present 15 \
    --present_rate 0.5

Edges appearing in fewer than 50% of seeds are discarded, yielding a stable consensus network.

Step 4 — Extract regulon candidate modules

scRBP getModule \
    --input              grn_consensus.tsv \
    --output_merged      modules.tsv \
    --importance_threshold 0.005 \
    --top_n_list         "5,10,50" \
    --target_top_n       "50" \
    --percentile         "0.75,0.9"

Step 5 — Prune with motif-binding evidence

scRBP getPrune \
    --rbp_targets        modules.tsv \
    --motif_rbp_links    motif2rbp.csv \
    --motif_target_ranks rankings.feather \
    --save_dir           ./pruned/ \
    --rank_threshold     1500

Step 6 — Export regulons to GMT format

scRBP getRegulon \
    --input       pruned/ctx_scores.csv \
    --out-symbol  regulons_symbol.gmt \
    --out-entrez  regulons_entrez.gmt \
    --map-custom  NCBI38.gene.loc \
    --min_genes   5

Step 7 — Merge region-specific GMT files

scRBP mergeRegulons \
    --base_dir ./analysis/ \
    --input    regulons_symbol.gmt \
    --output   regulons_combined.gmt \
    --recursive

Step 8 — Compute Regulon Activity Scores (RAS)

Uses the AUCell algorithm to score each cell or cell type for regulon activity. Also computes the Jensen–Shannon divergence-based Regulon Specificity Score (RSS).

scRBP ras \
    --mode         ct \
    --matrix       PBMC_sketch_15K.feather \
    --regulons     regulons_symbol.gmt \
    --out          ras_output/ \
    --celltypes-csv cell_to_celltype.csv

Step 9 — Regulon genetic association score (RGS)

Links each regulon to GWAS traits using MAGMA gene-set analysis with a 4D null distribution for empirical p-values.

scRBP rgs \
    --mode      ct \
    --magma     ~/tools/magma/magma \
    --genes-raw gwas.genes.raw \
    --sets      regulons_entrez.gmt \
    --id-type   entrez \
    --out       rgs_output/rgs

Step 10 — Compute Trait relevance Score (TRS)

Integrates RAS and RGS into a unified score:

TRS = norm(RAS) + norm(RGS) − λ × |norm(RAS) − norm(RGS)|

RBPs with high TRS are both activity-high in the cell type and genetically linked to the trait.

scRBP trs \
    --mode       ct \
    --ras        ras_output/aucell_ct.csv \
    --rgs_csv    rgs_output/rgs_real.csv \
    --out_prefix trs_output/trs

---

Command Reference

Step	Command	Key Inputs	Key Output
1	scRBP getSketch	.h5ad / .feather	Downsampled cells
2	scRBP getGRN	Expression matrix, RBP list	*_scRBP_gene_GRNs.tsv or *_scRBP_isoform_GRNs.tsv
3	scRBP getMerge_GRN	Multiple GRN TSV files (glob)	Consensus GRN TSV
4	scRBP getModule	Consensus GRN TSV	Modules TSV
5	scRBP getPrune	Modules TSV, motif files	Pruned scores (CSV)
6	scRBP getRegulon	Pruned scores	Regulons GMT (symbol + Entrez)
7	scRBP mergeRegulons	Multiple GMT files	Merged GMT
8	scRBP ras	Expression matrix, GMT	AUCell scores, RSS matrix
9	scRBP rgs	MAGMA .genes.raw, GMT	RGS scores CSV
10	scRBP trs	RAS CSV, RGS CSV	TRS scores CSV

Use scRBP <command> --help to see all parameters for any step.

---

Dependencies

Category	Packages
Core numerics	numpy, pandas, scipy, scikit-learn
Single-cell I/O	anndata, scanpy, loompy
Fast I/O	polars, pyarrow
Cell downsampling	geosketch
GRN inference	arboreto (GRNBoost2 / GENIE3)
Motif enrichment	ctxcore, pyscenic
Progress display	tqdm
GWAS enrichment	MAGMA binary (external, user-provided)

---

Citation

If you use scRBP in your research, please cite:

Ma Y. et al. Decoding disease-associated RNA-binding protein-mediated regulatory networks through polygenic enrichment across diverse cellular contexts. (2026)

---

License

MIT License. See LICENSE for details.

---

Links

• GitHub: https://github.com/mayunlong89/scRBP
• Issues: https://github.com/mayunlong89/scRBP/issues
• Full documentation: see scRBP_readme.md in the repository