ZebraCHOP

CRISPR guide-RNA design for zebrafish (Danio rerio), with Ensembl-aware post-processing and a clean web UI.

ZebraCHOP is a zebrafish-focused fork of CHOPCHOP. It picks high-scoring Cas9/TALEN/Cpf1/Nickase guides for any zebrafish gene, scores them with the published efficiency models (Doench 2016, Xu 2015, …), maps off-targets with Bowtie, and post-processes the output via Ensembl REST to bias selection toward early exons (more likely to produce loss-of-function).

---

What it does

   Gene name(s)
       │
       ▼
┌────────────────┐   ┌────────────────┐   ┌────────────────┐   ┌────────────────┐
│ Resolve via    │ → │ Extract        │ → │ Score guides   │ → │ Filter via     │
│ danRer11.gene  │   │ sequence with  │   │ + off-targets  │   │ Ensembl exon   │
│ _table         │   │ twoBitToFa     │   │ via Bowtie     │   │ structure      │
└────────────────┘   └────────────────┘   └────────────────┘   └────────────────┘
       │                                                              │
       └──────────────────────────► Per-gene .tsv (ranked) ◄──────────┘

Inputs: gene name(s) (or a genePred file). Outputs: one TSV per gene with ranked guides, off-target counts, GC%, self-complementarity, efficiency score, etc., plus an optional Ensembl-aware text summary.

---

Quick start

git clone https://github.com/abachu2005/ZebraCHOP.git
cd ZebraCHOP
python3 bin/zebrachop-setup     # interactive: detects bowtie/twoBitToFa/primer3, writes config_local.json
bash frontend/run.sh            # open http://127.0.0.1:8000

If you prefer the CLI:

python2 chopchop_query.py --gene_names rx3,tbx16 -o results/ -- -G danRer11
python  reformat.py -input results/ -output ranked_summary.txt

Why two Python versions?

The core CHOPCHOP engine (chopchop.py, chopchop_query.py, featurization.py) is Python 2.7 — it's a fork of upstream CHOPCHOP and depends on a frozen scikit-learn pickle (Doench_2016_18.01_model_nopos.pickle) that doesn't survive a clean Py3 port. The new web frontend and setup wizard are Python 3 and shell out to the Py2 engine as a subprocess.

The setup wizard records your Python-2 path under config_local.json["PYTHON2"] so you only have to set it once.

---

External dependencies

The setup wizard checks for these and helps you install/configure them:

Tool	Why
bowtie	Off-target alignment against the zebrafish genome
twoBitToFa (UCSC kent utils)	Extract genomic sequence around target
primer3 (optional)	Primer design around the chosen guide
danRer11.2bit	UCSC two-bit genome file (~770 MB, auto-downloadable)
Bowtie index	Built from the zebrafish FASTA via bowtie-build

You'll also want the included genetable/danRer11.gene_table (~4.8 MB) which the wizard already points at.

---

Repository layout

.
├── chopchop.py                  # Py2: core guide-design engine
├── chopchop_query.py            # Py2: batch wrapper, one TSV per gene
├── reformat.py                  # Py3 OK: Ensembl REST exon-aware post-processor
├── featurization.py             # Py2: ML feature builder (Doench 2016)
├── config.json                  # default tool paths (empty; copy to config_local.json)
├── frontend/                    # NEW: Python-3 FastAPI web UI (subprocesses chopchop_query.py)
│   ├── main.py
│   ├── index.html
│   ├── requirements.txt
│   └── run.sh
├── bin/zebrachop-setup          # NEW: Python-3 interactive setup wizard
├── genetable/danRer11.gene_table
├── results/                     # example output TSVs (rx3, tbx16, tbxta)
├── ZebraCHOP Command Line Batch Analysis Pipeline.pdf
└── LICENSE

Configuration (config_local.json)

The wizard writes this automatically. Manual example:

{
  "PATH": {
    "BOWTIE": "/usr/local/bin/bowtie",
    "TWOBITTOFA": "/usr/local/bin/twoBitToFa",
    "PRIMER3": "/usr/local/bin/primer3_core",
    "TWOBIT_INDEX_DIR": "/data/zebrachop/indexes",
    "BOWTIE_INDEX_DIR":  "/data/zebrachop/indexes",
    "GENE_TABLE_INDEX_DIR": "./genetable"
  },
  "THREADS": 4,
  "PYTHON2": "/usr/local/bin/python2"
}

config_local.json is git-ignored.

Web UI

Run bash frontend/run.sh and open http://127.0.0.1:8000. Paste one or more gene names, pick the scoring model, click Design guides — the frontend spawns a Python-2 subprocess per batch, streams a job log, and renders each gene's ranked TSV as a sortable table you can download.

Citing

If you use ZebraCHOP, please cite the upstream CHOPCHOP paper (Labun et al., NAR 2019). The poison-exon-aware Ensembl post-processing in reformat.py is an addition specific to this fork.

License

MIT — see LICENSE. Portions of the code derive from CHOPCHOP (MIT) and from Microsoft's Azimuth (BSD 3-Clause); see NOTICE and in-file headers for attributions.