Regulatr

Cell-type-aware non-coding variant prioritization.

Give it a single ENCODE chromatin-accessibility experiment and it tells you which disease-associated non-coding variants land in open regulatory DNA in that specific cell type — then maps them to their likely target genes and diseases and ranks them as mechanistic hypotheses.

Python standard library only (no third-party packages). All data is fetched live from public genomics resources at query time.

---

Why

Most disease-associated genetic variation is non-coding, where its effect — if any — is regulatory: it changes whether and where a gene is switched on. But regulatory DNA is cell-type-specific: an enhancer variant only matters in a cell type where that enhancer is active. So the useful question isn't "is this variant functional?" but:

"Is this variant in open, active regulatory DNA in the relevant cell type, and which gene does it plausibly regulate?"

regulatr uses chromatin accessibility (ATAC-seq / DNase peaks from ENCODE) as the readout of where regulatory DNA is open in a given biosample, and intersects that with known variants from ClinVar. Swap the accession for a different cell type and different loci rise to the top — that contrast is the point.

Scope guard: accessibility tells us where regulatory DNA is open; it never calls variants. Variants come from ClinVar. The single internal coordinate standard is hg38 / GRCh38, and build mismatches are rejected rather than silently lifted.

---

Features

• Accession resolution — enter an ENCODE experiment (ENCSR…) or peak file (ENCFF…); the tool picks the best GRCh38 peak file and records full provenance (assay, biosample, lab).
• Region discovery — scans the peak file genome-wide and surfaces the loci where this experiment has the strongest open chromatin, each labelled by its nearest gene.
• Variant retrieval — every ClinVar record in the region, with clinical significance, rsID, gnomAD allele frequency, and ref/alt.
• Accessibility overlap — marks which variants fall inside an open-chromatin peak in this cell type, and the peak's signal.
• Gene mapping — assigns each variant to its nearest gene via Ensembl.
• Disease inference — each target gene's single strongest Open Targets association (disease is inferred from the locus, never typed).
• Composite ranking — accessibility strength + ClinVar significance + Open Targets gene score, equally weighted; in-peak variants sort to the top.
• Pathway enrichment — an ordered g:Profiler query (GO:BP / Reactome / KEGG) over the cell type's most-accessible genes.

• Genome-browser view — embedded igv.js (hg38) with a self-contained SVG fallback.
• Export — full result set as JSON or CSV.

---

Quick start

cd app
./run.sh                 # http://127.0.0.1:8765   (or PORT=9000 ./run.sh)

No third-party packages — Python 3.8+ stdlib only.

Try ENCSR637OPZ (CD8 T cell) or ENCSR000EMT (GM12878 / B-cell): hit Load regions for each and compare which loci top the list. Deep-linkable: /?accession=ENCSR637OPZ&region=chr1:109200000-109350000 loads regions, then auto-runs.

---

How it works

For a single accession + region the backend composes seven deterministic tools:

resolve_accession → fetch_peaks → fetch_variants → overlap
                 → map_to_gene → opentargets_score → rank

Each variant's composite score is the mean of three terms in [0, 1]: accessibility (peak signal normalized to the strongest peak in the region), clinvar_pathogenicity (a weight from ClinVar significance), and opentargets_gene_score (the target gene's strongest association). In-peak status is the primary sort key, so cell-type accessibility dominates ordering.

The first call for a given accession downloads and caches its peak bed.gz under app/.cache/; later calls reuse it.

---

Project structure

app/
  server.py          # stdlib HTTP server + the Layer-1 tools and pipeline
  run.sh             # launch script
  static/            # index.html, app.js, style.css (frontend)
  README.md          # detailed backend/pipeline notes
LICENSE              # MIT

Data sources

ENCODE REST · NCBI ClinVar E-utilities · Ensembl REST · Open Targets GraphQL · g:Profiler. Genome build: hg38 / GRCh38 throughout.

---

Limitations

• Accessibility ≠ causality. A variant in an open peak is a hypothesis, not a proven regulatory effect.
• Nearest-gene mapping is a heuristic — not chromatin-contact evidence; enhancers can regulate distal genes.
• Coverage is bounded by the inputs — ClinVar is capped per region, and the gene-label pass during region discovery has a fetch budget, so the weakest hotspots may be coordinate-only.

---

License

MIT © 2026 Tarun Naithani