LOOM — Open-Science CRISPR Target Discovery

A 195 KB WebAssembly search engine that scans 700,000+ pathogen genomes in sub-millisecond time, entirely in the browser. No backend, no install, no API key.

Try the live tool →

---

What This Is

LOOM is a BWT/FM-index exact-retrieval engine built in Rust and compiled to WebAssembly. Its primary application is open-science CRISPR diagnostic target discovery: finding conserved 23-mer sequences across massive pathogen genome collections that could serve as diagnostic detection targets.

The project produced:

1. A browser-based search tool — select a pathogen, browse pre-computed targets, filter by gene/PAM/novelty, or run live FM-index search against the full genome collection
2. 52 novel CRISPR diagnostic targets — in genes with no prior published CRISPR diagnostics, verified through a 4-strategy literature scan with ontology-backed synonym expansion
3. A reproducible pipeline — every result in the paper can be regenerated from the scripts and data in this repository

Key Numbers

Metric	Value
WASM binary size	195 KB
Search latency	< 1 ms
Pathogen genomes indexed	12.3M+
Pre-computed targets	140,000+
Novel targets discovered	52
Off-target animal genomes screened	7 (human, mouse, cow, pig, chicken, bat, camel)

---

Repository Structure

brenda/                 Rust FM-index crate (the core engine)
  src/indexer/          Index construction: BWT, suffix arrays, sharding
  src/search/           Query engine: exact match, federated search

src/                    CLI + WASM bindings
  main.rs              Command-line interface entry point
  wasm.rs              WebAssembly bindings for browser deployment
  dna/                 Genomic pipeline: CRISPR scanning, off-target, conservation
  cli/                 CLI subcommands

web/                    Browser search interface (deployed to Azure Static Web Apps)
  crispr-search.html   Main search page — pathogen targets database + live search
  crispr-animal.html   Animal genome off-target screening
  learn.html           Interactive CRISPR learning guide
  methodology.html     v4 scan methodology & integrity report
  data/                Pre-computed JSON datasets served to the browser

data/crispr_guides/     Research results & validation data
  novel_52_guides.tsv   The 52 novel diagnostic targets (primary result)
  offtargets_*.csv      Off-target screening against 7 animal genomes
  patent_scan_report.json   Prior art screening results

scripts/                Reproducibility pipeline
  overnight_openscience.sh  Full pipeline orchestrator
  scan_crispr_targets.py    Target scanning wrapper
  compute_*.py/sh           Scoring, conservation, cross-reactivity, RNAfold
  pubmed_scan_v3.py         4-strategy literature gap verification (PubMed + Europe PMC)
  patent_scan.py            Lens.org prior art screening
  verify-citations.py       Citation verification for the paper
  ontology/                 Ontology fetchers (NCBI taxonomy, Disease Ontology, MONDO)

publications/           Research papers
  papers/               Source markdown for the paper
  preprints/            Built PDFs

docs/
  runbooks/             Pipeline operation guides
  plans/                Research planning documents
  lessons/              Development lessons learned

Getting Started

Prerequisites

• Rust (stable, 2024 edition)
• wasm-pack (for WASM builds)
• Python 3.10+ with a virtual environment (for pipeline scripts)

Build the CLI

cargo build --release

Build the WASM module

wasm-pack build --target web --out-dir pkg --release

Run the search tool locally

Open web/crispr-search.html in a browser, or serve the web/ directory:

cd web && python -m http.server 8080

Run the pipeline

See docs/runbooks/overnight-openscience-pipeline.md for full pipeline instructions.

# Scan targets for a pathogen
cargo run --release -- crispr-scan --fasta /path/to/genomes.fasta --output targets.json

# Score and filter guides
python scripts/compute_guide_scores.py

# Run off-target screening
bash scripts/compute_cross_reactivity.sh

CLI Commands

Command	Description
index	Build an FM-index from text or FASTA files
search	Exact-match search against an index
batch-search	Search multiple patterns in one pass
crispr-scan	Scan genomes for CRISPR-targetable 23-mers
guide-conservation	Compute conservation scores across genome sets
shard-search	Distributed search across sharded indexes

The Paper

The research paper describing the methodology and results is in publications/papers/:

• Full paper: pangenomic-crispr-targets-merged.md
• 5-page brief: pangenomic-crispr-targets-brief.md

An annotated walkthrough of the results is available at the live website.

License

MIT — see web/LICENSE

Author

Alvaro Videla Godoy — built with a Mac Studio and GitHub Copilot, evenings and weekends.