bio-clip-seq-stamp-antibody-free

What this skill does

## Version Compatibility

Reference examples tested with: STAMP / scSTAMP (Brannan 2021/2024 Yeo lab github), Bullseye 1.0+, SAILOR 1.1+, samtools 1.19+, REDItools2 1.3+, JACUSA2 2.0+, scanpy 1.10+, anndata 0.10+, pysam 0.22+.

Before using code patterns, verify installed versions match. If versions differ:
- Python: `pip show <package>` then `help(module.function)` to check signatures
- CLI: `<tool> --version` then `<tool> --help` to confirm flags

If code throws unexpected errors, introspect the installed package and adapt the example to match the actual CLI rather than retrying.

# STAMP / Antibody-Free RBP Profiling

**"Profile RBP-RNA targets without UV crosslinking or immunoprecipitation"** -> Express a fusion of the RBP-of-interest with a deaminase (APOBEC1 for STAMP, ADAR for TRIBE) in cells; the deaminase edits RNA nucleotides adjacent to where the RBP binds, producing a C-to-U (STAMP) or A-to-I (TRIBE, read as A-to-G) editing signature in standard RNA-seq. The targets are recovered computationally from the editing pattern. Three properties make this approach valuable: (a) no UV crosslinking required (works in tissue/in vivo); (b) no IP step (no antibody needed - the RBP itself targets the deaminase); (c) compatible with single-cell readout because the editing signal exists in standard scRNA-seq (scSTAMP, scTRIBE). Trade-off: editing is offset from the binding site (typically 0-50 nt away); resolution is approximate; off-target editing from deaminase alone must be subtracted.

- CLI (STAMP, bulk): standard RNA-seq pipeline + Bullseye or SAILOR for C-to-U edit detection vs APOBEC1-only control
- CLI (TRIBE, bulk): standard RNA-seq + REDItools2 or JACUSA2 for A-to-I edit detection vs ADAR-only control
- CLI (DART-seq for m6A): same as STAMP, with APOBEC1-YTH fusion (YTH is the m6A reader)
- Python (scSTAMP single-cell): 10x Genomics or Smart-seq2 pipeline + custom editing-rate quantification per cell + per-cell binding-target inference
- CLI (general edit-site detection): `JACUSA2 call-2 -r ref.fa -p 8 -F 1024 -A,B treated.bam,control.bam -t pileup.tsv` then filter for C-to-U or A-to-I

STAMP (Brannan 2021) is the canonical antibody-free RBP profiling method. TRIBE (McMahon 2016) and HyperTRIBE (Xu 2018) are earlier ADAR-based variants. DART-seq (Meyer 2019) is the m6A-specific application using YTH-fused APOBEC1. scSTAMP (Brannan 2024) extends STAMP to single-cell readout.

## Methods Taxonomy

| Method | Deaminase | Edit signature | Cells supported | Single-cell | Strength | Fails when |
|--------|-----------|----------------|-----------------|-------------|----------|------------|
| STAMP (Brannan 2021) | APOBEC1 | C->U in mRNA (reads as C->T) | Any | Yes (scSTAMP) | Antibody-free; no UV; in vivo | APOBEC1 also edits ssDNA off-target; saturated edits at high APOBEC1 expression |
| scSTAMP (Brannan 2024) | APOBEC1 | C->U per cell | Single cell (10x or Smart-seq2) | Yes (native) | Per-cell RBP profiling | Coverage per cell limits sensitivity; ~25% of cytosines accessible per transcript |
| TRIBE (McMahon 2016) | ADAR catalytic domain | A->I (reads as A->G in cDNA) | Drosophila standard; mammalian works | Yes (scTRIBE) | First antibody-free | Edits restricted to certain ADAR consensus; lower edit rate than HyperTRIBE |
| HyperTRIBE (Xu 2018) | ADAR E488Q hyperactive mutant | A->I in much wider context | Drosophila / mammalian | Yes | Higher edit rate than original TRIBE | Hyperactive may edit off-target; needs ADAR-only control |
| TRIBE-DiCo (Erickson 2024) | ADAR catalytic + dimerization | A->I | Mammalian | Yes | Improved specificity | Newer; less validation |
| DART-seq (Meyer 2019) | APOBEC1-YTH | C->U near m6A | Any | Yes (scDART) | m6A reader profiling | Indirect (edits near m6A, not at RBP binding sites) |
| Bullseye | NA (analysis tool) | NA | Any | Yes | STAMP / DART analysis | Just an analysis pipeline |
| SAILOR | NA (analysis tool) | NA | Any | Yes | RNA editing analysis | Just an analysis pipeline |
| REDItools2 | NA (analysis tool) | NA | Any | NA | Generic RNA editing | Generic; not RBP-specific |
| JACUSA2 (Piechotta 2022 BMC Bioinformatics 23:139) | NA (analysis tool) | NA | Any | NA | Multi-sample edit-site detection | Generic; not RBP-specific |
| ADAR-CLIP | NA - this is regular ADAR CLIP | NA | NA | NA | CLIP for ADAR | Not an antibody-free method; just a different CLIP target |

Methodology evolves; the Brannan lab and Yeo lab papers (2021, 2024) are canonical. Verify deaminase fusion expression level (low expression for specificity; saturation degrades specificity).

## STAMP vs m6A-Specific Methods

For m6A profiling specifically, antibody-free choices include:
- **DART-seq (APOBEC1-YTH fusion):** This skill covers the methodology, but for m6A detection see clip-seq/m6a-clip. Only 44% of DART edits fall within DRACH motifs (Liu 2023); strong off-target component.
- **GLORI (Liu 2023):** Antibody-free, chemical, stoichiometric single-base m6A; this is the new (2023) gold standard for m6A. See clip-seq/m6a-clip.
- **m6Anet (Hendra 2022):** Nanopore direct RNA m6A; AUC 0.83 on HEK293T.

If the use case is m6A profiling, the m6a-clip skill is the canonical reference; this skill (stamp-antibody-free) focuses on the broader RBP-editing-fusion paradigm where the target is not m6A but the RBP's RNA targets.

## Critical Choice: STAMP (APOBEC1) vs TRIBE (ADAR)

| Property | STAMP | TRIBE |
|----------|-------|-------|
| Deaminase | APOBEC1 (cytidine -> uridine) | ADAR (adenosine -> inosine) |
| Edit signature | C->U (reads as C->T) | A->I (reads as A->G) |
| ssRNA preference | Yes (APOBEC1 acts on ssRNA + ssDNA) | No (ADAR acts on dsRNA stems by default; ADAR2cd in TRIBE relaxes this) |
| Edit clusters per target | 10-1000 | ~5-50 (lower; HyperTRIBE higher) |
| Off-target | APOBEC1 alone has detectable C->U on ssDNA + RNA | ADAR has weak intrinsic A->I |
| Spatial offset from RBP binding | 0-50 nt | 0-30 nt |
| Cell line tested | HEK293, K562, mouse tissue | Drosophila (original), mouse, human |
| Single-cell | scSTAMP 2024 | scTRIBE 2020 |
| Compatible methods | C->U is rare in mRNA; signal is clean | A->I is common at ALU repeats; baseline ADAR editing competes |
| Cytosine accessibility | ~25-35% of mRNA bases are C; APOBEC1 needs ssRNA | All A residues are potential ADAR targets |

Both work; STAMP has more clusters per target (advantage for low-coverage scenarios) and cleaner background (C->U is rare in mRNA). TRIBE has more flexibility (ADAR variants tunable) and lower off-target. Practical choice often comes down to lab familiarity.

## scSTAMP / scTRIBE Single-Cell Workflow

The defining advantage of antibody-free RBP profiling is compatibility with single-cell readout. scSTAMP processes 10x Genomics or Smart-seq2 libraries.

```bash
# Standard 10x cellranger pipeline produces BAM with per-cell barcodes
cellranger count \
    --id=scstamp_sample \
    --transcriptome=refdata-gex-GRCh38 \
    --fastqs=fastq_dir \
    --localcores=16 --localmem=64

# scSTAMP analysis (Yeo lab github)
# Quantify per-cell C->U editing
python scstamp_analysis.py \
    --bam scstamp_sample/outs/possorted_genome_bam.bam \
    --barcodes scstamp_sample/outs/filtered_feature_bc_matrix/barcodes.tsv.gz \
    --control apobec1_only_sample/outs/possorted_genome_bam.bam \
    --output per_cell_edits.h5
```

Per-cell edit-rate matrix can be integrated with standard scRNA-seq clustering. The per-cell binding profile is reconstructed from cells with sufficient coverage (>= 10000 unique reads typically).

## Editing-Site Detection Pipelines

**Goal:** Recover specific (not background) RBP-fusion-induced editing sites from RNA-seq libraries by subtracting the deaminase-only control.

**Approach:** Process fusion-sample BAM and deaminase-only-control BAM in parallel; use Bullseye (STAMP/DART), SAILOR (Yeo), or JACUSA2 (general) to call C-to-U (STAMP/DART) or A-to-I (TRIBE/HyperTRIBE) edit sites at edit rate >= 0.1 and coverage >= 10, requiring fusion-vs